diff --git a/.gitignore b/.gitignore
index 9bd28491..93f79906 100644
--- a/.gitignore
+++ b/.gitignore
@@ -18,5 +18,7 @@
 
 # Build artifacts
 bin
+/.bin/
+/kubevpn
 
 *.DS_Store
diff --git a/README.md b/README.md
index 43f804e0..984690e0 100644
--- a/README.md
+++ b/README.md
@@ -247,6 +247,36 @@ as `productpage`
 ***Disclaimer:*** This only works on the namespace where kubevpn-traffic-manager is deployed. Otherwise,
 use [Domain resolve](./README.md#domain-resolve)
 
+### Local outbound proxy for nested VPNs
+
+If your operating system routes are controlled by another VPN client, route-based access may not work reliably even though `kubevpn connect` succeeds. In that case, use a local outbound proxy instead of depending on cluster CIDR routing.
+
+Start a standalone local SOCKS5 proxy:
+
+```shell
+➜  ~ kubevpn proxy-out --listen-socks 127.0.0.1:1080
+Proxy hint: export ALL_PROXY=socks5h://127.0.0.1:1080
+Proxy scope: TCP traffic only. Use socks5h if you want proxy-side cluster DNS resolution.
+```
+
+Then access cluster Services, Pod IPs, or Service ClusterIPs through the proxy:
+
+```shell
+➜  ~ curl --proxy socks5h://127.0.0.1:1080 http://productpage.default.svc.cluster.local:9080
+➜  ~ curl --proxy socks5h://127.0.0.1:1080 http://172.21.10.49:9080
+```
+
+Or start the managed SOCKS5 proxy as part of `connect`:
+
+```shell
+➜  ~ kubevpn connect --socks
+Started managed local SOCKS5 proxy for connection 03dc50feb8c3 on 127.0.0.1:1080
+Proxy hint: export ALL_PROXY=socks5h://127.0.0.1:1080
+Proxy scope: TCP traffic only. Use socks5h if you want proxy-side cluster DNS resolution.
+```
+
+Use `kubevpn disconnect <connection-id>` or `kubevpn quit` to stop the managed proxy. In this version, outbound proxy mode supports TCP traffic only.
+
 ### Connect to multiple kubernetes cluster network
 
 already connected cluster `ccijorbccotmqodvr189g`
@@ -756,4 +786,4 @@ If you want to debug this project on local PC. Please follow the steps bellow:
 
 [![JetBrains logo.](https://resources.jetbrains.com/storage/products/company/brand/logos/jetbrains.svg)](https://jb.gg/OpenSourceSupport)
 
-### [Donate](https://kubevpn.dev/docs/donate)
\ No newline at end of file
+### [Donate](https://kubevpn.dev/docs/donate)
diff --git a/README_ZH.md b/README_ZH.md
index 69be1402..fc779cb8 100644
--- a/README_ZH.md
+++ b/README_ZH.md
@@ -234,6 +234,36 @@ reviews                   ClusterIP   172.21.8.24     <none>        9080/TCP
 
 可以看到直接使用 service name 的方式，可以正常访问到集群资源。
 
+### 嵌套 VPN 场景下的本地出站代理
+
+如果本机路由被另外一个 VPN 客户端接管，那么即使 `kubevpn connect` 成功，基于路由的访问也可能不稳定。这种情况下可以改用本地出站代理，而不是依赖集群 CIDR 路由。
+
+启动一个独立的本地 SOCKS5 代理：
+
+```shell
+➜  ~ kubevpn proxy-out --listen-socks 127.0.0.1:1080
+Proxy hint: export ALL_PROXY=socks5h://127.0.0.1:1080
+Proxy scope: TCP traffic only. Use socks5h if you want proxy-side cluster DNS resolution.
+```
+
+然后通过代理访问集群中的 Service、Pod IP 或 Service ClusterIP：
+
+```shell
+➜  ~ curl --proxy socks5h://127.0.0.1:1080 http://productpage.default.svc.cluster.local:9080
+➜  ~ curl --proxy socks5h://127.0.0.1:1080 http://172.21.10.49:9080
+```
+
+也可以在 `connect` 时自动启动托管的 SOCKS5 代理：
+
+```shell
+➜  ~ kubevpn connect --socks
+Started managed local SOCKS5 proxy for connection 03dc50feb8c3 on 127.0.0.1:1080
+Proxy hint: export ALL_PROXY=socks5h://127.0.0.1:1080
+Proxy scope: TCP traffic only. Use socks5h if you want proxy-side cluster DNS resolution.
+```
+
+可以通过 `kubevpn disconnect <connection-id>` 或 `kubevpn quit` 停止这个托管代理。当前版本只支持 TCP 代理流量。
+
 ### 链接到多集群网络
 
 可以看到已经链接到了一个集群 `ccijorbccotmqodvr189g`
diff --git a/cmd/kubevpn/cmds/connect.go b/cmd/kubevpn/cmds/connect.go
index abeefa92..a3c47250 100644
--- a/cmd/kubevpn/cmds/connect.go
+++ b/cmd/kubevpn/cmds/connect.go
@@ -2,7 +2,9 @@ package cmds
 
 import (
 	"context"
+	"errors"
 	"fmt"
+	"io"
 	"os"
 
 	log "github.com/sirupsen/logrus"
@@ -29,6 +31,8 @@ func CmdConnect(f cmdutil.Factory) *cobra.Command {
 	var extraRoute = &handler.ExtraRouteInfo{}
 	var sshConf = &pkgssh.SshConfig{}
 	var transferImage, foreground bool
+	var enableSocks bool
+	var socksListen string
 	var imagePullSecretName string
 	var managerNamespace string
 	cmd := &cobra.Command{
@@ -46,6 +50,10 @@ func CmdConnect(f cmdutil.Factory) *cobra.Command {
 		# Connect to k8s cluster network
 		kubevpn connect
 
+		# Connect and start a managed local SOCKS5 proxy for nested VPN cases
+		kubevpn connect --socks
+		curl --proxy socks5h://127.0.0.1:1080 http://productpage.default.svc.cluster.local:9080
+
 		# Connect to api-server behind of bastion host or ssh jump host
 		kubevpn connect --ssh-addr 192.168.1.100:22 --ssh-username root --ssh-keyfile ~/.ssh/ssh.pem
 
@@ -115,13 +123,22 @@ func CmdConnect(f cmdutil.Factory) *cobra.Command {
 			if err != nil {
 				return err
 			}
-			err = util.PrintGRPCStream[rpc.ConnectResponse](cmd.Context(), resp)
+			connectionID, err := printConnectGRPCStream(cmd.Context(), resp, os.Stdout)
 			if err != nil {
 				if status.Code(err) == codes.Canceled {
 					return nil
 				}
 				return err
 			}
+			if enableSocks {
+				if connectionID == "" {
+					return fmt.Errorf("connected but missing connection ID for socks proxy startup")
+				}
+				if err := startManagedSocksProxy(connectionID, bytes, ns, socksListen); err != nil {
+					return err
+				}
+				printManagedSocksStarted(os.Stdout, connectionID, socksListen)
+			}
 			if !foreground {
 				_, _ = fmt.Fprintln(os.Stdout, config.Slogan)
 			} else {
@@ -137,6 +154,8 @@ func CmdConnect(f cmdutil.Factory) *cobra.Command {
 	}
 	handler.AddCommonFlags(cmd.Flags(), &transferImage, &imagePullSecretName)
 	cmd.Flags().BoolVar(&foreground, "foreground", false, "Hang up")
+	cmd.Flags().BoolVar(&enableSocks, "socks", false, "Start a managed local SOCKS5 proxy after connect for nested VPN or route-conflict cases")
+	cmd.Flags().StringVar(&socksListen, "socks-listen", "127.0.0.1:1080", "Listen address for the managed local SOCKS5 proxy")
 	cmd.Flags().StringVar(&managerNamespace, "manager-namespace", "", "The namespace where the traffic manager is to be found. Only works in cluster mode (install kubevpn server by helm)")
 
 	handler.AddExtraRoute(cmd.Flags(), extraRoute)
@@ -144,7 +163,35 @@ func CmdConnect(f cmdutil.Factory) *cobra.Command {
 	return cmd
 }
 
+func printConnectGRPCStream(ctx context.Context, clientStream grpc.ClientStream, out io.Writer) (string, error) {
+	go func() {
+		if ctx != nil {
+			<-ctx.Done()
+			_ = clientStream.SendMsg(&rpc.Cancel{})
+		}
+	}()
+
+	var connectionID string
+	for {
+		resp := new(rpc.ConnectResponse)
+		err := clientStream.RecvMsg(resp)
+		if errors.Is(err, io.EOF) {
+			return connectionID, nil
+		}
+		if err != nil {
+			return connectionID, err
+		}
+		if resp.ConnectionID != "" {
+			connectionID = resp.ConnectionID
+		}
+		if out != nil && resp.GetMessage() != "" {
+			_, _ = fmt.Fprint(out, resp.GetMessage())
+		}
+	}
+}
+
 func disconnect(cli rpc.DaemonClient, bytes []byte, ns string, sshConf *pkgssh.SshConfig) error {
+	connectionID, _ := connectionIDFromKubeconfigBytes(bytes, ns)
 	resp, err := cli.Disconnect(context.Background())
 	if err != nil {
 		plog.G(context.Background()).Errorf("Disconnect error: %v", err)
@@ -166,5 +213,24 @@ func disconnect(cli rpc.DaemonClient, bytes []byte, ns string, sshConf *pkgssh.S
 		}
 		return err
 	}
+	if connectionID != "" {
+		_ = stopManagedProxy(connectionID)
+		printManagedSocksStopped(os.Stdout, connectionID)
+	}
 	return nil
 }
+
+func connectionIDFromKubeconfigBytes(bytes []byte, ns string) (string, error) {
+	file, err := util.ConvertToTempKubeconfigFile(bytes, "")
+	if err != nil {
+		return "", err
+	}
+	defer os.Remove(file)
+
+	factory := util.InitFactoryByPath(file, ns)
+	connect := &handler.ConnectOptions{}
+	if err := connect.InitClient(factory); err != nil {
+		return "", err
+	}
+	return util.GetConnectionID(context.Background(), connect.GetClientset().CoreV1().Namespaces(), connect.Namespace)
+}
diff --git a/cmd/kubevpn/cmds/connect_test.go b/cmd/kubevpn/cmds/connect_test.go
new file mode 100644
index 00000000..400fd15a
--- /dev/null
+++ b/cmd/kubevpn/cmds/connect_test.go
@@ -0,0 +1,97 @@
+package cmds
+
+import (
+	"bytes"
+	"context"
+	"io"
+	"testing"
+	"time"
+
+	"google.golang.org/grpc/metadata"
+
+	"github.com/wencaiwulue/kubevpn/v2/pkg/daemon/rpc"
+)
+
+type fakeConnectClientStream struct {
+	ctx        context.Context
+	responses  []*rpc.ConnectResponse
+	recvIndex  int
+	cancelSent int
+	waitOnDone bool
+}
+
+func (f *fakeConnectClientStream) Header() (metadata.MD, error) { return nil, nil }
+func (f *fakeConnectClientStream) Trailer() metadata.MD         { return nil }
+func (f *fakeConnectClientStream) CloseSend() error             { return nil }
+func (f *fakeConnectClientStream) Context() context.Context {
+	if f.ctx != nil {
+		return f.ctx
+	}
+	return context.Background()
+}
+func (f *fakeConnectClientStream) SendMsg(m interface{}) error {
+	if _, ok := m.(*rpc.Cancel); ok {
+		f.cancelSent++
+	}
+	return nil
+}
+func (f *fakeConnectClientStream) RecvMsg(m interface{}) error {
+	if f.recvIndex >= len(f.responses) {
+		if f.waitOnDone && f.ctx != nil {
+			<-f.ctx.Done()
+			time.Sleep(10 * time.Millisecond)
+		}
+		return io.EOF
+	}
+	resp, ok := m.(*rpc.ConnectResponse)
+	if !ok {
+		return io.ErrUnexpectedEOF
+	}
+	*resp = *f.responses[f.recvIndex]
+	f.recvIndex++
+	return nil
+}
+
+func TestPrintConnectGRPCStreamReturnsConnectionID(t *testing.T) {
+	stream := &fakeConnectClientStream{
+		responses: []*rpc.ConnectResponse{
+			{Message: "Starting connect\n"},
+			{ConnectionID: "conn-123", Message: "Connected tunnel\n"},
+		},
+	}
+	var out bytes.Buffer
+
+	connectionID, err := printConnectGRPCStream(context.Background(), stream, &out)
+	if err != nil {
+		t.Fatal(err)
+	}
+	if connectionID != "conn-123" {
+		t.Fatalf("connectionID = %q, want %q", connectionID, "conn-123")
+	}
+	if got := out.String(); got != "Starting connect\nConnected tunnel\n" {
+		t.Fatalf("unexpected output %q", got)
+	}
+}
+
+func TestPrintConnectGRPCStreamSendsCancelOnContextDone(t *testing.T) {
+	ctx, cancel := context.WithCancel(context.Background())
+	stream := &fakeConnectClientStream{ctx: ctx, waitOnDone: true}
+
+	done := make(chan struct{})
+	go func() {
+		defer close(done)
+		_, _ = printConnectGRPCStream(ctx, stream, io.Discard)
+	}()
+
+	cancel()
+
+	select {
+	case <-done:
+	case <-time.After(2 * time.Second):
+		t.Fatal("printConnectGRPCStream did not finish after cancel")
+	}
+
+	if stream.cancelSent == 0 {
+		t.Fatal("expected cancel message to be sent to the grpc stream")
+	}
+}
diff --git a/cmd/kubevpn/cmds/disconnect.go b/cmd/kubevpn/cmds/disconnect.go
index 4aa44251..b05b2888 100644
--- a/cmd/kubevpn/cmds/disconnect.go
+++ b/cmd/kubevpn/cmds/disconnect.go
@@ -76,6 +76,13 @@ func CmdDisconnect(f cmdutil.Factory) *cobra.Command {
 				}
 				return err
 			}
+			if all {
+				_ = stopAllManagedProxies()
+				_, _ = fmt.Fprintln(os.Stdout, "Stopped all managed local SOCKS proxies")
+			} else if id != "" {
+				_ = stopManagedProxy(id)
+				printManagedSocksStopped(os.Stdout, id)
+			}
 			_, _ = fmt.Fprint(os.Stdout, "Disconnect completed")
 			return nil
 		},
diff --git a/cmd/kubevpn/cmds/proxy_out.go b/cmd/kubevpn/cmds/proxy_out.go
new file mode 100644
index 00000000..99503cb3
--- /dev/null
+++ b/cmd/kubevpn/cmds/proxy_out.go
@@ -0,0 +1,124 @@
+package cmds
+
+import (
+	"fmt"
+	"io"
+	"os"
+
+	"github.com/spf13/cobra"
+	cmdutil "k8s.io/kubectl/pkg/cmd/util"
+	"k8s.io/kubectl/pkg/util/i18n"
+	"k8s.io/kubectl/pkg/util/templates"
+
+	"github.com/wencaiwulue/kubevpn/v2/pkg/localproxy"
+	plog "github.com/wencaiwulue/kubevpn/v2/pkg/log"
+)
+
+func CmdProxyOut(f cmdutil.Factory) *cobra.Command {
+	var socksListen string
+	var httpConnectListen string
+
+	cmd := &cobra.Command{
+		Use:   "proxy-out",
+		Short: i18n.T("Expose cluster workloads through a local outbound proxy"),
+		Long: templates.LongDesc(i18n.T(`
+Expose cluster workloads through a local outbound proxy.
+
+This command is designed for nested VPN cases where OS routes are owned by
+another VPN client. Instead of depending on cluster CIDR routing, it resolves
+cluster service hostnames and forwards TCP traffic through the Kubernetes API.
+Use SOCKS5 with socks5h when you want the proxy to resolve cluster DNS names.
+`)),
+		Example: templates.Examples(i18n.T(`
+		# Start a local SOCKS5 proxy and access a cluster Service through socks5h
+		kubevpn proxy-out --listen-socks 127.0.0.1:1080
+		curl --proxy socks5h://127.0.0.1:1080 http://productpage.default.svc.cluster.local:9080
+
+		# Start both SOCKS5 and HTTP CONNECT listeners for TCP traffic
+		kubevpn proxy-out --listen-socks 127.0.0.1:1080 --listen-http-connect 127.0.0.1:3128
+
+		# Access a Pod IP or Service ClusterIP through the proxy
+		curl --proxy socks5h://127.0.0.1:1080 http://172.21.10.49:9080
+		`)),
+		RunE: func(cmd *cobra.Command, args []string) error {
+			plog.InitLoggerForClient()
+
+			if err := validateProxyOutListeners(socksListen, httpConnectListen); err != nil {
+				return err
+			}
+
+			restConfig, err := f.ToRESTConfig()
+			if err != nil {
+				return err
+			}
+			clusterAPI, clientset, err := localproxy.NewClusterAPI(restConfig)
+			if err != nil {
+				return err
+			}
+			ns, _, err := f.ToRawKubeConfigLoader().Namespace()
+			if err != nil {
+				return err
+			}
+
+			connector := &localproxy.ClusterConnector{
+				Client:           clusterAPI,
+				Forwarder:        localproxy.NewPodDialer(restConfig, clientset),
+				RESTConfig:       restConfig,
+				DefaultNamespace: ns,
+			}
+			server := &localproxy.Server{
+				Connector:         connector,
+				SOCKSListenAddr:   socksListen,
+				HTTPConnectListen: httpConnectListen,
+				Stdout:            os.Stdout,
+				Stderr:            os.Stderr,
+			}
+
+			if ip := localproxy.FirstNonLoopbackIPv4(); ip != "" {
+				_, _ = fmt.Fprintf(os.Stdout, "Local host IPv4 detected: %s\n", ip)
+			}
+			printProxyOutHints(os.Stdout, socksListen, httpConnectListen)
+			return server.ListenAndServe(cmd.Context())
+		},
+	}
+
+	cmd.Flags().StringVar(&socksListen, "listen-socks", "127.0.0.1:1080", "Local SOCKS5 listen address")
+	cmd.Flags().StringVar(&httpConnectListen, "listen-http-connect", "", "Local HTTP CONNECT listen address")
+	return cmd
+}
+
+func validateProxyOutListeners(socksListen, httpConnectListen string) error {
+	if socksListen == "" && httpConnectListen == "" {
+		return fmt.Errorf("at least one of --listen-socks or --listen-http-connect must be set")
+	}
+	return nil
+}
+
+func printManagedSocksStarted(out io.Writer, connectionID, socksListen string) {
+	if out == nil {
+		return
+	}
+	_, _ = fmt.Fprintf(out, "Started managed local SOCKS5 proxy for connection %s on %s\n", connectionID, socksListen)
+	printProxyOutHints(out, socksListen, "")
+}
+
+func printManagedSocksStopped(out io.Writer, connectionID string) {
+	if out == nil || connectionID == "" {
+		return
+	}
+	_, _ = fmt.Fprintf(out, "Stopped managed local SOCKS5 proxy for connection %s\n", connectionID)
+}
+
+func printProxyOutHints(out io.Writer, socksListen, httpConnectListen string) {
+	if out == nil {
+		return
+	}
+	if socksListen != "" {
+		_, _ = fmt.Fprintf(out, "Proxy hint: export ALL_PROXY=socks5h://%s\n", socksListen)
+	}
+	if httpConnectListen != "" {
+		_, _ = fmt.Fprintf(out, "Proxy hint: export HTTP_PROXY=http://%s\n", httpConnectListen)
+		_, _ = fmt.Fprintf(out, "Proxy hint: export HTTPS_PROXY=http://%s\n", httpConnectListen)
+	}
+	_, _ = fmt.Fprintln(out, "Proxy scope: TCP traffic only. Use socks5h if you want proxy-side cluster DNS resolution.")
+}
diff --git a/cmd/kubevpn/cmds/proxy_out_manager.go b/cmd/kubevpn/cmds/proxy_out_manager.go
new file mode 100644
index 00000000..23d7f9d2
--- /dev/null
+++ b/cmd/kubevpn/cmds/proxy_out_manager.go
@@ -0,0 +1,179 @@
+package cmds
+
+import (
+	"encoding/json"
+	"fmt"
+	"os"
+	"os/exec"
+	"path/filepath"
+	"time"
+
+	"github.com/wencaiwulue/kubevpn/v2/pkg/config"
+	"github.com/wencaiwulue/kubevpn/v2/pkg/util"
+)
+
+type managedProxyInfo struct {
+	ConnectionID   string    `json:"connectionID"`
+	PID            int       `json:"pid"`
+	SocksAddress   string    `json:"socksAddress"`
+	KubeconfigPath string    `json:"kubeconfigPath"`
+	Namespace      string    `json:"namespace"`
+	LogPath        string    `json:"logPath"`
+	StartedAt      time.Time `json:"startedAt"`
+}
+
+var execProxyOutCommand = exec.Command
+var resolveCurrentExecutable = os.Executable
+var proxyOutStateDirFunc = func() string {
+	return filepath.Join(filepath.Dir(config.GetDBPath()), "proxy-out")
+}
+var proxyOutLogPathFunc = func(connectionID string) string {
+	return filepath.Join(filepath.Dir(config.GetDaemonLogPath(false)), fmt.Sprintf("proxy-out-%s.log", connectionID))
+}
+
+func proxyOutStateDir() string {
+	return proxyOutStateDirFunc()
+}
+
+func proxyOutStatePath(connectionID string) string {
+	return filepath.Join(proxyOutStateDir(), connectionID+".json")
+}
+
+func proxyOutLogPath(connectionID string) string {
+	return proxyOutLogPathFunc(connectionID)
+}
+
+func ensureProxyOutStateDir() error {
+	return os.MkdirAll(proxyOutStateDir(), 0755)
+}
+
+func startManagedSocksProxy(connectionID string, kubeconfigBytes []byte, namespace, listenAddr string) error {
+	if connectionID == "" {
+		return fmt.Errorf("connection ID is required")
+	}
+	if listenAddr == "" {
+		listenAddr = "127.0.0.1:1080"
+	}
+	if err := ensureProxyOutStateDir(); err != nil {
+		return err
+	}
+	_ = stopManagedProxy(connectionID)
+
+	kubeconfigPath := filepath.Join(config.GetTempPath(), fmt.Sprintf("proxy-out-%s.kubeconfig", connectionID))
+	if _, err := util.ConvertToTempKubeconfigFile(kubeconfigBytes, kubeconfigPath); err != nil {
+		return err
+	}
+
+	logPath := proxyOutLogPath(connectionID)
+	if err := os.MkdirAll(filepath.Dir(logPath), 0755); err != nil {
+		_ = os.Remove(kubeconfigPath)
+		return err
+	}
+	logFile, err := os.OpenFile(logPath, os.O_CREATE|os.O_WRONLY|os.O_APPEND, 0644)
+	if err != nil {
+		_ = os.Remove(kubeconfigPath)
+		return err
+	}
+	defer logFile.Close()
+
+	exe, err := resolveCurrentExecutable()
+	if err != nil {
+		return err
+	}
+
+	args := []string{
+		"--kubeconfig", kubeconfigPath,
+		"--namespace", namespace,
+		"proxy-out",
+		"--listen-socks", listenAddr,
+	}
+	cmd := execProxyOutCommand(exe, args...)
+	cmd.Stdout = logFile
+	cmd.Stderr = logFile
+	cmd.Stdin = nil
+	cmd.Env = os.Environ()
+	detachProcess(cmd)
+
+	if err := cmd.Start(); err != nil {
+		_ = os.Remove(kubeconfigPath)
+		return err
+	}
+
+	info := managedProxyInfo{
+		ConnectionID:   connectionID,
+		PID:            cmd.Process.Pid,
+		SocksAddress:   listenAddr,
+		KubeconfigPath: kubeconfigPath,
+		Namespace:      namespace,
+		LogPath:        logPath,
+		StartedAt:      time.Now(),
+	}
+	if err := writeManagedProxyInfo(info); err != nil {
+		_ = cmd.Process.Kill()
+		_ = os.Remove(kubeconfigPath)
+		return err
+	}
+	return nil
+}
+
+func writeManagedProxyInfo(info managedProxyInfo) error {
+	if err := ensureProxyOutStateDir(); err != nil {
+		return err
+	}
+	data, err := json.Marshal(info)
+	if err != nil {
+		return err
+	}
+	return os.WriteFile(proxyOutStatePath(info.ConnectionID), data, 0644)
+}
+
+func readManagedProxyInfo(connectionID string) (*managedProxyInfo, error) {
+	data, err := os.ReadFile(proxyOutStatePath(connectionID))
+	if err != nil {
+		return nil, err
+	}
+	var info managedProxyInfo
+	if err := json.Unmarshal(data, &info); err != nil {
+		return nil, err
+	}
+	return &info, nil
+}
+
+func stopManagedProxy(connectionID string) error {
+	info, err := readManagedProxyInfo(connectionID)
+	if err != nil {
+		if os.IsNotExist(err) {
+			return nil
+		}
+		return err
+	}
+	if info.PID > 0 {
+		process, err := os.FindProcess(info.PID)
+		if err == nil {
+			_ = process.Kill()
+		}
+	}
+	if info.KubeconfigPath != "" {
+		_ = os.Remove(info.KubeconfigPath)
+	}
+	_ = os.Remove(proxyOutStatePath(connectionID))
+	return nil
+}
+
+func stopAllManagedProxies() error {
+	entries, err := os.ReadDir(proxyOutStateDir())
+	if err != nil {
+		if os.IsNotExist(err) {
+			return nil
+		}
+		return err
+	}
+	for _, entry := range entries {
+		if entry.IsDir() || filepath.Ext(entry.Name()) != ".json" {
+			continue
+		}
+		connectionID := entry.Name()[:len(entry.Name())-len(".json")]
+		_ = stopManagedProxy(connectionID)
+	}
+	return nil
+}
diff --git a/cmd/kubevpn/cmds/proxy_out_manager_test.go b/cmd/kubevpn/cmds/proxy_out_manager_test.go
new file mode 100644
index 00000000..00acdb20
--- /dev/null
+++ b/cmd/kubevpn/cmds/proxy_out_manager_test.go
@@ -0,0 +1,155 @@
+//go:build !windows
+
+package cmds
+
+import (
+	"os"
+	"os/exec"
+	"path/filepath"
+	"testing"
+
+	"github.com/wencaiwulue/kubevpn/v2/pkg/util"
+)
+
+func TestManagedProxyInfoRoundTrip(t *testing.T) {
+	tempDir := t.TempDir()
+	oldStateDir := proxyOutStateDirFunc
+	oldLogPath := proxyOutLogPathFunc
+	proxyOutStateDirFunc = func() string { return tempDir }
+	proxyOutLogPathFunc = func(connectionID string) string { return filepath.Join(tempDir, connectionID+".log") }
+	defer func() {
+		proxyOutStateDirFunc = oldStateDir
+		proxyOutLogPathFunc = oldLogPath
+	}()
+
+	info := managedProxyInfo{
+		ConnectionID: "abc123",
+		PID:          12345,
+		SocksAddress: "127.0.0.1:1080",
+		Namespace:    "kubevpn",
+	}
+	if err := writeManagedProxyInfo(info); err != nil {
+		t.Fatal(err)
+	}
+	got, err := readManagedProxyInfo("abc123")
+	if err != nil {
+		t.Fatal(err)
+	}
+	if got.ConnectionID != info.ConnectionID || got.PID != info.PID || got.SocksAddress != info.SocksAddress {
+		t.Fatalf("unexpected managed proxy info: %#v", got)
+	}
+}
+
+func TestStopManagedProxyRemovesStateAndKubeconfig(t *testing.T) {
+	tempDir := t.TempDir()
+	oldStateDir := proxyOutStateDirFunc
+	oldLogPath := proxyOutLogPathFunc
+	proxyOutStateDirFunc = func() string { return tempDir }
+	proxyOutLogPathFunc = func(connectionID string) string { return filepath.Join(tempDir, connectionID+".log") }
+	defer func() {
+		proxyOutStateDirFunc = oldStateDir
+		proxyOutLogPathFunc = oldLogPath
+	}()
+
+	kubeconfigPath := filepath.Join(tempDir, "proxy.kubeconfig")
+	if _, err := util.ConvertToTempKubeconfigFile([]byte(`{"apiVersion":"v1","kind":"Config","clusters":[],"contexts":[],"users":[]}`), kubeconfigPath); err != nil {
+		t.Fatal(err)
+	}
+	if err := writeManagedProxyInfo(managedProxyInfo{
+		ConnectionID:   "conn1",
+		PID:            999999,
+		KubeconfigPath: kubeconfigPath,
+	}); err != nil {
+		t.Fatal(err)
+	}
+
+	if err := stopManagedProxy("conn1"); err != nil {
+		t.Fatal(err)
+	}
+	if _, err := os.Stat(proxyOutStatePath("conn1")); !os.IsNotExist(err) {
+		t.Fatalf("expected state file removed, got err=%v", err)
+	}
+	if _, err := os.Stat(kubeconfigPath); !os.IsNotExist(err) {
+		t.Fatalf("expected kubeconfig removed, got err=%v", err)
+	}
+}
+
+func TestStartManagedSocksProxyWritesState(t *testing.T) {
+	tempDir := t.TempDir()
+	oldStateDir := proxyOutStateDirFunc
+	oldLogPath := proxyOutLogPathFunc
+	oldExec := execProxyOutCommand
+	oldResolve := resolveCurrentExecutable
+	proxyOutStateDirFunc = func() string { return tempDir }
+	proxyOutLogPathFunc = func(connectionID string) string { return filepath.Join(tempDir, connectionID+".log") }
+	resolveCurrentExecutable = func() (string, error) { return "/bin/sh", nil }
+	execProxyOutCommand = func(name string, args ...string) *exec.Cmd {
+		return exec.Command("/bin/sh", "-c", "sleep 30")
+	}
+	defer func() {
+		proxyOutStateDirFunc = oldStateDir
+		proxyOutLogPathFunc = oldLogPath
+		execProxyOutCommand = oldExec
+		resolveCurrentExecutable = oldResolve
+	}()
+
+	if err := startManagedSocksProxy("conn2", []byte(`{"apiVersion":"v1","kind":"Config","clusters":[],"contexts":[],"users":[]}`), "kubevpn", "127.0.0.1:1089"); err != nil {
+		t.Fatal(err)
+	}
+	info, err := readManagedProxyInfo("conn2")
+	if err != nil {
+		t.Fatal(err)
+	}
+	if info.PID <= 0 || info.SocksAddress != "127.0.0.1:1089" {
+		t.Fatalf("unexpected managed proxy info: %#v", info)
+	}
+	_ = stopManagedProxy("conn2")
+}
+
+func TestStartManagedSocksProxyReplacesPreviousState(t *testing.T) {
+	tempDir := t.TempDir()
+	oldStateDir := proxyOutStateDirFunc
+	oldLogPath := proxyOutLogPathFunc
+	oldExec := execProxyOutCommand
+	oldResolve := resolveCurrentExecutable
+	proxyOutStateDirFunc = func() string { return tempDir }
+	proxyOutLogPathFunc = func(connectionID string) string { return filepath.Join(tempDir, connectionID+".log") }
+	resolveCurrentExecutable = func() (string, error) { return "/bin/sh", nil }
+	execProxyOutCommand = func(name string, args ...string) *exec.Cmd {
+		return exec.Command("/bin/sh", "-c", "sleep 30")
+	}
+	defer func() {
+		proxyOutStateDirFunc = oldStateDir
+		proxyOutLogPathFunc = oldLogPath
+		execProxyOutCommand = oldExec
+		resolveCurrentExecutable = oldResolve
+	}()
+
+	staleKubeconfig := filepath.Join(tempDir, "stale.kubeconfig")
+	if _, err := util.ConvertToTempKubeconfigFile([]byte(`{"apiVersion":"v1","kind":"Config","clusters":[],"contexts":[],"users":[]}`), staleKubeconfig); err != nil {
+		t.Fatal(err)
+	}
+	if err := writeManagedProxyInfo(managedProxyInfo{
+		ConnectionID:   "conn-restart",
+		PID:            999999,
+		KubeconfigPath: staleKubeconfig,
+	}); err != nil {
+		t.Fatal(err)
+	}
+
+	if err := startManagedSocksProxy("conn-restart", []byte(`{"apiVersion":"v1","kind":"Config","clusters":[],"contexts":[],"users":[]}`), "kubevpn", "127.0.0.1:1090"); err != nil {
+		t.Fatal(err)
+	}
+
+	if _, err := os.Stat(staleKubeconfig); !os.IsNotExist(err) {
+		t.Fatalf("expected stale kubeconfig removed, got err=%v", err)
+	}
+	info, err := readManagedProxyInfo("conn-restart")
+	if err != nil {
+		t.Fatal(err)
+	}
+	if info.PID <= 0 || info.SocksAddress != "127.0.0.1:1090" {
+		t.Fatalf("unexpected managed proxy info after restart: %#v", info)
+	}
+	_ = stopManagedProxy("conn-restart")
+}
diff --git a/cmd/kubevpn/cmds/proxy_out_manager_unix.go b/cmd/kubevpn/cmds/proxy_out_manager_unix.go
new file mode 100644
index 00000000..0660524d
--- /dev/null
+++ b/cmd/kubevpn/cmds/proxy_out_manager_unix.go
@@ -0,0 +1,12 @@
+//go:build !windows
+
+package cmds
+
+import (
+	"os/exec"
+	"syscall"
+)
+
+func detachProcess(cmd *exec.Cmd) {
+	cmd.SysProcAttr = &syscall.SysProcAttr{Setpgid: true}
+}
diff --git a/cmd/kubevpn/cmds/proxy_out_manager_windows.go b/cmd/kubevpn/cmds/proxy_out_manager_windows.go
new file mode 100644
index 00000000..c1462f8a
--- /dev/null
+++ b/cmd/kubevpn/cmds/proxy_out_manager_windows.go
@@ -0,0 +1,7 @@
+//go:build windows
+
+package cmds
+
+import "os/exec"
+
+func detachProcess(cmd *exec.Cmd) {}
diff --git a/cmd/kubevpn/cmds/proxy_out_test.go b/cmd/kubevpn/cmds/proxy_out_test.go
new file mode 100644
index 00000000..0a4649bd
--- /dev/null
+++ b/cmd/kubevpn/cmds/proxy_out_test.go
@@ -0,0 +1,37 @@
+package cmds
+
+import (
+	"bytes"
+	"strings"
+	"testing"
+)
+
+func TestValidateProxyOutListeners(t *testing.T) {
+	if err := validateProxyOutListeners("", ""); err == nil {
+		t.Fatal("expected validation error when no listeners are configured")
+	}
+	if err := validateProxyOutListeners("127.0.0.1:1080", ""); err != nil {
+		t.Fatalf("unexpected error for socks listener: %v", err)
+	}
+	if err := validateProxyOutListeners("", "127.0.0.1:3128"); err != nil {
+		t.Fatalf("unexpected error for http connect listener: %v", err)
+	}
+}
+
+func TestPrintProxyOutHints(t *testing.T) {
+	var out bytes.Buffer
+	printProxyOutHints(&out, "127.0.0.1:1080", "127.0.0.1:3128")
+
+	got := out.String()
+	for _, want := range []string{
+		"ALL_PROXY=socks5h://127.0.0.1:1080",
+		"HTTP_PROXY=http://127.0.0.1:3128",
+		"HTTPS_PROXY=http://127.0.0.1:3128",
+		"TCP traffic only",
+		"socks5h",
+	} {
+		if !strings.Contains(got, want) {
+			t.Fatalf("expected output to contain %q, got %q", want, got)
+		}
+	}
+}
diff --git a/cmd/kubevpn/cmds/quit.go b/cmd/kubevpn/cmds/quit.go
index c9f4fdf8..055877d5 100644
--- a/cmd/kubevpn/cmds/quit.go
+++ b/cmd/kubevpn/cmds/quit.go
@@ -31,6 +31,8 @@ func CmdQuit(f cmdutil.Factory) *cobra.Command {
 		RunE: func(cmd *cobra.Command, args []string) error {
 			_ = quit(cmd.Context(), true)
 			_ = quit(cmd.Context(), false)
+			_ = stopAllManagedProxies()
+			_, _ = fmt.Fprintln(os.Stdout, "Stopped all managed local SOCKS proxies")
 			util.CleanExtensionLib()
 			_, _ = fmt.Fprint(os.Stdout, "Exited")
 			return nil
diff --git a/cmd/kubevpn/cmds/root.go b/cmd/kubevpn/cmds/root.go
index f4ff4309..28c40b62 100644
--- a/cmd/kubevpn/cmds/root.go
+++ b/cmd/kubevpn/cmds/root.go
@@ -57,6 +57,7 @@ func NewKubeVPNCommand() *cobra.Command {
 				CmdConnect(factory),
 				CmdDisconnect(factory),
 				CmdProxy(factory),
+				CmdProxyOut(factory),
 				CmdLeave(factory),
 				CmdSync(factory),
 				CmdUnsync(factory),
diff --git a/docs/en/LocalProxyDesign.md b/docs/en/LocalProxyDesign.md
new file mode 100644
index 00000000..694944b3
--- /dev/null
+++ b/docs/en/LocalProxyDesign.md
@@ -0,0 +1,136 @@
+# Local Proxy Design For Nested VPNs
+
+## Summary
+
+`proxy-out` adds a local outbound proxy mode for environments where `kubevpn connect` succeeds but operating-system route ownership is controlled by another VPN client. Instead of relying on cluster CIDR routing, client applications send TCP traffic to a local proxy, and kubevpn forwards that traffic into the cluster through the Kubernetes API.
+
+This version ships:
+
+- a standalone `kubevpn proxy-out` command
+- a managed SOCKS5 mode via `kubevpn connect --socks`
+- SOCKS5 and optional HTTP CONNECT listeners
+- remote resolution of cluster Service names when clients use `socks5h`
+- support for Service ClusterIP and Pod IP TCP targets
+
+This version does not ship:
+
+- PAC files
+- a DNS listener
+- transparent interception
+- `kubevpn status` integration for proxy listener state
+
+## Problem
+
+In nested VPN setups, another VPN client may own the host route table or DNS path. That breaks kubevpn's normal route-based access even when the tunnel itself is healthy. The most common symptoms are:
+
+- cluster IP traffic leaves through the outer tunnel instead of kubevpn
+- split DNS settings do not resolve cluster names end to end
+- local applications cannot reliably reach Services or Pod IPs without manual proxy support
+
+## Implemented Design
+
+### Data path
+
+For outbound proxy mode, the user-facing API is a local listener:
+
+```text
+local app
+  -> SOCKS5 or HTTP CONNECT
+  -> kubevpn proxy-out server
+  -> Kubernetes API port-forward to a resolved Pod endpoint
+  -> cluster destination
+```
+
+### Name and target resolution
+
+- Service hostnames such as `svc`, `svc.ns`, `svc.ns.svc.cluster.local` are resolved inside kubevpn
+- Service ClusterIP targets are mapped back to a Service, then to a ready endpoint Pod
+- Pod IP targets dial the matching running Pod directly
+- short names use the active kubeconfig namespace as the default namespace
+
+### Lifecycle
+
+- `kubevpn proxy-out` runs as a foreground local proxy process
+- `kubevpn connect --socks` starts a managed background SOCKS5 proxy bound to the connection ID
+- `kubevpn disconnect <connection-id>`, `kubevpn disconnect --all`, and `kubevpn quit` stop managed proxies
+- managed state is stored under the kubevpn local state directory; transient kubeconfig files are deleted on stop
+- proxy logs are intentionally retained for troubleshooting
+
+## User Experience
+
+### Standalone mode
+
+```bash
+kubevpn proxy-out --listen-socks 127.0.0.1:1080
+curl --proxy socks5h://127.0.0.1:1080 http://productpage.default.svc.cluster.local:9080
+```
+
+### Managed mode
+
+```bash
+kubevpn connect --socks
+curl --proxy socks5h://127.0.0.1:1080 http://productpage.default.svc.cluster.local:9080
+```
+
+### Notes
+
+- default SOCKS5 listen address is `127.0.0.1:1080`
+- HTTP CONNECT is optional and disabled by default
+- proxy mode handles TCP traffic only in this version
+- use `socks5h` when you want proxy-side cluster DNS resolution
+
+## Implementation Notes
+
+- Command entrypoints live in [/Users/tvorogme/projects/kubevpn/cmd/kubevpn/cmds/proxy_out.go](/Users/tvorogme/projects/kubevpn/cmd/kubevpn/cmds/proxy_out.go) and [/Users/tvorogme/projects/kubevpn/cmd/kubevpn/cmds/connect.go](/Users/tvorogme/projects/kubevpn/cmd/kubevpn/cmds/connect.go)
+- Managed proxy state handling lives in [/Users/tvorogme/projects/kubevpn/cmd/kubevpn/cmds/proxy_out_manager.go](/Users/tvorogme/projects/kubevpn/cmd/kubevpn/cmds/proxy_out_manager.go)
+- Local proxy implementation lives in [/Users/tvorogme/projects/kubevpn/pkg/localproxy](/Users/tvorogme/projects/kubevpn/pkg/localproxy)
+
+## Validation
+
+Recommended validation scenarios:
+
+1. Start `kubevpn proxy-out --listen-socks 127.0.0.1:1080`
+2. Access a Service hostname through `socks5h`
+3. Access a Service ClusterIP through `socks5h`
+4. Start `kubevpn connect --socks`
+5. Confirm `kubevpn disconnect` or `kubevpn quit` removes the managed proxy state and kubeconfig file
+
+### Milestone 1
+
+- design doc
+- CLI scaffolding
+- daemon action scaffolding
+- in-memory SOCKS5 server using current connection state
+
+### Milestone 2
+
+- real resolver/dialer integration
+- status output
+- documentation and examples
+
+### Milestone 3
+
+- HTTP CONNECT
+- optional PAC file generation
+- optional DNS helper listener
+
+## Why This Is The Right Feature
+
+Today kubevpn is strongest when it can own system routing. Nested VPN environments break that assumption.
+
+A local explicit proxy:
+
+- avoids fighting other VPN route owners
+- gives users an app-level integration surface
+- reuses kubevpn's existing transport and remote DNS
+- solves the exact class of failures seen with `sing-box` packet tunnel on macOS
+
+## Next Implementation Slice
+
+The smallest useful next coding step is:
+
+1. add `kubevpn proxy-out` command and flags
+2. add daemon action to attach proxy mode to an existing connection
+3. implement TCP-only SOCKS5 CONNECT
+4. implement hostname resolution through kubevpn remote DNS
+5. verify `curl --proxy socks5h://127.0.0.1:1080` against a cluster service
diff --git a/pkg/handler/connect.go b/pkg/handler/connect.go
index f3ff0f25..8e712eab 100644
--- a/pkg/handler/connect.go
+++ b/pkg/handler/connect.go
@@ -1218,7 +1218,9 @@ func healthCheckPortForward(ctx context.Context, cancelFunc context.CancelFunc,
 	}
 
 	var healthChecker = func() error {
-		conn, err := net.Dial("tcp", fmt.Sprintf(":%s", localGvisorUDPPort))
+		// Use loopback explicitly so the health check keeps working even when the
+		// default route is owned by another tunnel.
+		conn, err := net.Dial("tcp", net.JoinHostPort("127.0.0.1", localGvisorUDPPort))
 		if err != nil {
 			return err
 		}
diff --git a/pkg/localproxy/cluster.go b/pkg/localproxy/cluster.go
new file mode 100644
index 00000000..0403a9c0
--- /dev/null
+++ b/pkg/localproxy/cluster.go
@@ -0,0 +1,271 @@
+package localproxy
+
+import (
+	"context"
+	"fmt"
+	"net"
+	"strings"
+
+	corev1 "k8s.io/api/core/v1"
+	metav1 "k8s.io/apimachinery/pkg/apis/meta/v1"
+	"k8s.io/apimachinery/pkg/fields"
+	"k8s.io/client-go/kubernetes"
+	"k8s.io/client-go/rest"
+)
+
+type ClusterConnector struct {
+	Client           ClusterAPI
+	Forwarder        PodDialer
+	RESTConfig       *rest.Config
+	DefaultNamespace string
+}
+
+type resolvedTarget struct {
+	PodName   string
+	Namespace string
+	PodPort   int32
+}
+
+type ClusterAPI interface {
+	GetService(ctx context.Context, namespace, name string) (*corev1.Service, error)
+	GetEndpoints(ctx context.Context, namespace, name string) (*corev1.Endpoints, error)
+	ListServices(ctx context.Context) (*corev1.ServiceList, error)
+	ListPodsByIP(ctx context.Context, ip string) (*corev1.PodList, error)
+}
+
+type PodDialer interface {
+	DialPod(ctx context.Context, namespace, podName string, port int32) (net.Conn, error)
+}
+
+type kubeClusterAPI struct {
+	clientset kubernetes.Interface
+}
+
+func NewClusterAPI(config *rest.Config) (ClusterAPI, *kubernetes.Clientset, error) {
+	clientset, err := kubernetes.NewForConfig(config)
+	if err != nil {
+		return nil, nil, err
+	}
+	return &kubeClusterAPI{clientset: clientset}, clientset, nil
+}
+
+func (k *kubeClusterAPI) GetService(ctx context.Context, namespace, name string) (*corev1.Service, error) {
+	return k.clientset.CoreV1().Services(namespace).Get(ctx, name, metav1.GetOptions{})
+}
+
+func (k *kubeClusterAPI) GetEndpoints(ctx context.Context, namespace, name string) (*corev1.Endpoints, error) {
+	return k.clientset.CoreV1().Endpoints(namespace).Get(ctx, name, metav1.GetOptions{})
+}
+
+func (k *kubeClusterAPI) ListServices(ctx context.Context) (*corev1.ServiceList, error) {
+	return k.clientset.CoreV1().Services("").List(ctx, metav1.ListOptions{})
+}
+
+func (k *kubeClusterAPI) ListPodsByIP(ctx context.Context, ip string) (*corev1.PodList, error) {
+	return k.clientset.CoreV1().Pods("").List(ctx, metav1.ListOptions{
+		FieldSelector: fields.OneTermEqualSelector("status.podIP", ip).String(),
+	})
+}
+
+func (c *ClusterConnector) Connect(ctx context.Context, host string, port int) (net.Conn, error) {
+	target, err := c.resolveTarget(ctx, host, port)
+	if err != nil {
+		return nil, err
+	}
+	if c.Forwarder == nil {
+		return nil, fmt.Errorf("pod dialer is required")
+	}
+	return c.Forwarder.DialPod(ctx, target.Namespace, target.PodName, target.PodPort)
+}
+
+func (c *ClusterConnector) resolveTarget(ctx context.Context, host string, port int) (*resolvedTarget, error) {
+	if ip := net.ParseIP(host); ip != nil {
+		if target, err := c.resolveServiceIP(ctx, host, port); err == nil {
+			return target, nil
+		}
+		return c.resolvePodIP(ctx, host, port)
+	}
+	return c.resolveServiceHost(ctx, host, port)
+}
+
+func (c *ClusterConnector) resolveServiceHost(ctx context.Context, host string, port int) (*resolvedTarget, error) {
+	svcName, ns, ok := parseServiceHost(host, c.DefaultNamespace)
+	if !ok {
+		return nil, fmt.Errorf("unsupported cluster host %q", host)
+	}
+	svc, err := c.Client.GetService(ctx, ns, svcName)
+	if err != nil {
+		return nil, err
+	}
+	return c.resolveServiceToPod(ctx, svc, port)
+}
+
+func (c *ClusterConnector) resolveServiceIP(ctx context.Context, host string, port int) (*resolvedTarget, error) {
+	services, err := c.Client.ListServices(ctx)
+	if err != nil {
+		return nil, err
+	}
+	for i := range services.Items {
+		if services.Items[i].Spec.ClusterIP == host {
+			return c.resolveServiceToPod(ctx, &services.Items[i], port)
+		}
+	}
+	return nil, fmt.Errorf("no service found for cluster ip %s", host)
+}
+
+func (c *ClusterConnector) resolveServiceToPod(ctx context.Context, svc *corev1.Service, port int) (*resolvedTarget, error) {
+	if svc == nil {
+		return nil, fmt.Errorf("service is required")
+	}
+	var servicePort *corev1.ServicePort
+	for i := range svc.Spec.Ports {
+		if int(svc.Spec.Ports[i].Port) == port {
+			servicePort = &svc.Spec.Ports[i]
+			break
+		}
+	}
+	if servicePort == nil {
+		if len(svc.Spec.Ports) == 1 {
+			servicePort = &svc.Spec.Ports[0]
+		} else {
+			return nil, fmt.Errorf("service %s/%s has no port %d", svc.Namespace, svc.Name, port)
+		}
+	}
+
+	endpoints, err := c.Client.GetEndpoints(ctx, svc.Namespace, svc.Name)
+	if err != nil {
+		return nil, err
+	}
+	var endpointPort *corev1.EndpointPort
+	var podName, podNS, podIP string
+	for _, subset := range endpoints.Subsets {
+		for i := range subset.Ports {
+			portRef := &subset.Ports[i]
+			if endpointPortMatches(servicePort, portRef, port) {
+				endpointPort = portRef
+				break
+			}
+		}
+		if endpointPort == nil && len(subset.Ports) == 1 {
+			endpointPort = &subset.Ports[0]
+		}
+		if endpointPort == nil {
+			continue
+		}
+		for _, addr := range subset.Addresses {
+			podIP = addr.IP
+			if addr.TargetRef != nil && addr.TargetRef.Kind == "Pod" {
+				podName = addr.TargetRef.Name
+				podNS = addr.TargetRef.Namespace
+				break
+			}
+		}
+		if podName != "" || podIP != "" {
+			break
+		}
+	}
+	if endpointPort == nil {
+		return nil, fmt.Errorf("service %s/%s has no ready endpoints for port %d", svc.Namespace, svc.Name, port)
+	}
+	if podName == "" {
+		if podIP == "" {
+			return nil, fmt.Errorf("service %s/%s has endpoints without pod references", svc.Namespace, svc.Name)
+		}
+		target, err := c.resolvePodIP(ctx, podIP, int(endpointPort.Port))
+		if err != nil {
+			return nil, err
+		}
+		target.PodPort = endpointPort.Port
+		return target, nil
+	}
+	if podNS == "" {
+		podNS = svc.Namespace
+	}
+	return &resolvedTarget{
+		PodName:   podName,
+		Namespace: podNS,
+		PodPort:   endpointPort.Port,
+	}, nil
+}
+
+func endpointPortMatches(servicePort *corev1.ServicePort, endpointPort *corev1.EndpointPort, requestedPort int) bool {
+	if servicePort == nil || endpointPort == nil {
+		return false
+	}
+	if servicePort.Name != "" && endpointPort.Name != "" && servicePort.Name == endpointPort.Name {
+		return true
+	}
+	if endpointPort.Port == int32(requestedPort) {
+		return true
+	}
+	if servicePort.TargetPort.IntValue() > 0 && int32(servicePort.TargetPort.IntValue()) == endpointPort.Port {
+		return true
+	}
+	return false
+}
+
+func (c *ClusterConnector) resolvePodIP(ctx context.Context, host string, port int) (*resolvedTarget, error) {
+	pods, err := c.Client.ListPodsByIP(ctx, host)
+	if err != nil {
+		return nil, err
+	}
+	for i := range pods.Items {
+		pod := &pods.Items[i]
+		if pod.Status.Phase != corev1.PodRunning || pod.DeletionTimestamp != nil {
+			continue
+		}
+		return &resolvedTarget{
+			PodName:   pod.Name,
+			Namespace: pod.Namespace,
+			PodPort:   int32(port),
+		}, nil
+	}
+	return nil, fmt.Errorf("no running pod found for ip %s", host)
+}
+
+func parseServiceHost(host, defaultNamespace string) (service string, namespace string, ok bool) {
+	host = strings.TrimSuffix(host, ".")
+	parts := strings.Split(host, ".")
+	switch {
+	case len(parts) >= 5 && parts[2] == "svc" && parts[3] == "cluster" && parts[4] == "local":
+		return parts[0], parts[1], true
+	case len(parts) >= 3 && parts[2] == "svc":
+		return parts[0], parts[1], true
+	case len(parts) == 2:
+		return parts[0], parts[1], true
+	case len(parts) == 1 && defaultNamespace != "":
+		return parts[0], defaultNamespace, true
+	default:
+		return "", "", false
+	}
+}
+
+func FirstNonLoopbackIPv4() string {
+	ifaces, err := net.Interfaces()
+	if err != nil {
+		return ""
+	}
+	for _, iface := range ifaces {
+		if iface.Flags&net.FlagUp == 0 || iface.Flags&net.FlagLoopback != 0 {
+			continue
+		}
+		addrs, err := iface.Addrs()
+		if err != nil {
+			continue
+		}
+		for _, addr := range addrs {
+			var ip net.IP
+			switch v := addr.(type) {
+			case *net.IPNet:
+				ip = v.IP
+			case *net.IPAddr:
+				ip = v.IP
+			}
+			if ip == nil || ip.IsLoopback() || ip.To4() == nil {
+				continue
+			}
+			return ip.String()
+		}
+	}
+	return ""
+}
diff --git a/pkg/localproxy/httpconnect.go b/pkg/localproxy/httpconnect.go
new file mode 100644
index 00000000..035d6445
--- /dev/null
+++ b/pkg/localproxy/httpconnect.go
@@ -0,0 +1,60 @@
+package localproxy
+
+import (
+	"bufio"
+	"context"
+	"fmt"
+	"net"
+	"net/http"
+	"strconv"
+)
+
+func ServeHTTPConnect(ctx context.Context, ln net.Listener, connector Connector) error {
+	for {
+		conn, err := ln.Accept()
+		if err != nil {
+			select {
+			case <-ctx.Done():
+				return nil
+			default:
+				return err
+			}
+		}
+		go handleHTTPConnectConn(ctx, conn, connector)
+	}
+}
+
+func handleHTTPConnectConn(ctx context.Context, conn net.Conn, connector Connector) {
+	defer conn.Close()
+
+	reader := bufio.NewReader(conn)
+	req, err := http.ReadRequest(reader)
+	if err != nil {
+		return
+	}
+	if req.Method != http.MethodConnect {
+		_, _ = fmt.Fprint(conn, "HTTP/1.1 405 Method Not Allowed\r\nConnection: close\r\n\r\n")
+		return
+	}
+
+	host, portStr, err := net.SplitHostPort(req.Host)
+	if err != nil {
+		_, _ = fmt.Fprint(conn, "HTTP/1.1 400 Bad Request\r\nConnection: close\r\n\r\n")
+		return
+	}
+	port, err := strconv.Atoi(portStr)
+	if err != nil {
+		_, _ = fmt.Fprint(conn, "HTTP/1.1 400 Bad Request\r\nConnection: close\r\n\r\n")
+		return
+	}
+
+	remote, err := connector.Connect(ctx, host, port)
+	if err != nil {
+		_, _ = fmt.Fprint(conn, "HTTP/1.1 502 Bad Gateway\r\nConnection: close\r\n\r\n")
+		return
+	}
+	defer remote.Close()
+
+	_, _ = fmt.Fprint(conn, "HTTP/1.1 200 Connection Established\r\n\r\n")
+	relayConns(conn, remote)
+}
diff --git a/pkg/localproxy/portforward.go b/pkg/localproxy/portforward.go
new file mode 100644
index 00000000..586850c8
--- /dev/null
+++ b/pkg/localproxy/portforward.go
@@ -0,0 +1,87 @@
+package localproxy
+
+import (
+	"context"
+	"fmt"
+	"io"
+	"net"
+	"strconv"
+	"sync"
+
+	"k8s.io/client-go/kubernetes"
+	"k8s.io/client-go/rest"
+
+	"github.com/wencaiwulue/kubevpn/v2/pkg/util"
+)
+
+type kubePortForwarder struct {
+	config     *rest.Config
+	restClient *rest.RESTClient
+}
+
+func NewPodDialer(config *rest.Config, clientset *kubernetes.Clientset) PodDialer {
+	restClient, _ := clientset.CoreV1().RESTClient().(*rest.RESTClient)
+	return &kubePortForwarder{
+		config:     config,
+		restClient: restClient,
+	}
+}
+
+func (k *kubePortForwarder) DialPod(ctx context.Context, namespace, podName string, port int32) (net.Conn, error) {
+	if k.restClient == nil {
+		return nil, fmt.Errorf("pod port-forward requires REST client")
+	}
+	listener, err := net.Listen("tcp", "127.0.0.1:0")
+	if err != nil {
+		return nil, err
+	}
+	localPort := listener.Addr().(*net.TCPAddr).Port
+	_ = listener.Close()
+
+	stopChan := make(chan struct{})
+	readyChan := make(chan struct{})
+	errCh := make(chan error, 1)
+	go func() {
+		errCh <- util.PortForwardPod(
+			k.config,
+			k.restClient,
+			podName,
+			namespace,
+			[]string{fmt.Sprintf("%d:%d", localPort, port)},
+			readyChan,
+			stopChan,
+			io.Discard,
+			io.Discard,
+		)
+	}()
+
+	select {
+	case <-readyChan:
+	case err := <-errCh:
+		return nil, err
+	case <-ctx.Done():
+		close(stopChan)
+		return nil, ctx.Err()
+	}
+
+	dialer := net.Dialer{}
+	conn, err := dialer.DialContext(ctx, "tcp", net.JoinHostPort("127.0.0.1", strconv.Itoa(localPort)))
+	if err != nil {
+		close(stopChan)
+		return nil, err
+	}
+	return &forwardedConn{Conn: conn, stop: stopChan}, nil
+}
+
+type forwardedConn struct {
+	net.Conn
+	stop chan struct{}
+	once sync.Once
+}
+
+func (c *forwardedConn) Close() error {
+	c.once.Do(func() {
+		close(c.stop)
+	})
+	return c.Conn.Close()
+}
diff --git a/pkg/localproxy/relay.go b/pkg/localproxy/relay.go
new file mode 100644
index 00000000..1d2b2e6e
--- /dev/null
+++ b/pkg/localproxy/relay.go
@@ -0,0 +1,23 @@
+package localproxy
+
+import (
+	"io"
+	"net"
+	"sync"
+)
+
+func relayConns(left, right net.Conn) {
+	var wg sync.WaitGroup
+	copyHalf := func(dst, src net.Conn) {
+		defer wg.Done()
+		_, _ = io.Copy(dst, src)
+		if tcp, ok := dst.(*net.TCPConn); ok {
+			_ = tcp.CloseWrite()
+		}
+	}
+
+	wg.Add(2)
+	go copyHalf(left, right)
+	go copyHalf(right, left)
+	wg.Wait()
+}
diff --git a/pkg/localproxy/server.go b/pkg/localproxy/server.go
new file mode 100644
index 00000000..b1ece36d
--- /dev/null
+++ b/pkg/localproxy/server.go
@@ -0,0 +1,103 @@
+package localproxy
+
+import (
+	"context"
+	"fmt"
+	"io"
+	"net"
+	"sync"
+)
+
+type Connector interface {
+	Connect(ctx context.Context, host string, port int) (net.Conn, error)
+}
+
+type Server struct {
+	Connector         Connector
+	SOCKSListenAddr   string
+	HTTPConnectListen string
+	Stdout            io.Writer
+	Stderr            io.Writer
+}
+
+func (s *Server) ListenAndServe(ctx context.Context) error {
+	if s.Connector == nil {
+		return fmt.Errorf("connector is required")
+	}
+
+	var listeners []net.Listener
+	closeAll := func() {
+		for _, ln := range listeners {
+			_ = ln.Close()
+		}
+	}
+	defer closeAll()
+
+	type serveFn func(context.Context, net.Listener, Connector) error
+	var entries []struct {
+		name  string
+		addr  string
+		serve serveFn
+	}
+	if s.SOCKSListenAddr != "" {
+		entries = append(entries, struct {
+			name  string
+			addr  string
+			serve serveFn
+		}{name: "SOCKS5", addr: s.SOCKSListenAddr, serve: ServeSOCKS5})
+	}
+	if s.HTTPConnectListen != "" {
+		entries = append(entries, struct {
+			name  string
+			addr  string
+			serve serveFn
+		}{name: "HTTP CONNECT", addr: s.HTTPConnectListen, serve: ServeHTTPConnect})
+	}
+
+	errCh := make(chan error, len(entries))
+	var wg sync.WaitGroup
+	for _, entry := range entries {
+		ln, err := net.Listen("tcp", entry.addr)
+		if err != nil {
+			return err
+		}
+		listeners = append(listeners, ln)
+		if s.Stdout != nil {
+			_, _ = fmt.Fprintf(s.Stdout, "%s proxy listening on %s\n", entry.name, ln.Addr().String())
+		}
+		wg.Add(1)
+		go func(serve serveFn, ln net.Listener) {
+			defer wg.Done()
+			if err := serve(ctx, ln, s.Connector); err != nil && !isListenerClosedError(err) {
+				errCh <- err
+			}
+		}(entry.serve, ln)
+	}
+
+	done := make(chan struct{})
+	go func() {
+		defer close(done)
+		wg.Wait()
+	}()
+
+	select {
+	case err := <-errCh:
+		return err
+	case <-ctx.Done():
+		closeAll()
+		<-done
+		return nil
+	case <-done:
+		return nil
+	}
+}
+
+func isListenerClosedError(err error) bool {
+	if err == nil {
+		return false
+	}
+	if oe, ok := err.(*net.OpError); ok && oe.Err != nil {
+		return oe.Err.Error() == "use of closed network connection"
+	}
+	return err.Error() == "use of closed network connection"
+}
diff --git a/pkg/localproxy/server_test.go b/pkg/localproxy/server_test.go
new file mode 100644
index 00000000..1086d87b
--- /dev/null
+++ b/pkg/localproxy/server_test.go
@@ -0,0 +1,435 @@
+package localproxy
+
+import (
+	"bufio"
+	"context"
+	"fmt"
+	"io"
+	"net"
+	"net/http"
+	"net/http/httptest"
+	"net/url"
+	"strings"
+	"sync"
+	"testing"
+	"time"
+
+	xproxy "golang.org/x/net/proxy"
+	corev1 "k8s.io/api/core/v1"
+	metav1 "k8s.io/apimachinery/pkg/apis/meta/v1"
+	"k8s.io/apimachinery/pkg/util/intstr"
+)
+
+type fakeConnector struct {
+	addr string
+
+	mu    sync.Mutex
+	hosts []string
+	ports []int
+}
+
+func (f *fakeConnector) Connect(ctx context.Context, host string, port int) (net.Conn, error) {
+	f.mu.Lock()
+	f.hosts = append(f.hosts, host)
+	f.ports = append(f.ports, port)
+	f.mu.Unlock()
+
+	var d net.Dialer
+	return d.DialContext(ctx, "tcp", f.addr)
+}
+
+func (f *fakeConnector) lastTarget() (string, int) {
+	f.mu.Lock()
+	defer f.mu.Unlock()
+	if len(f.hosts) == 0 {
+		return "", 0
+	}
+	return f.hosts[len(f.hosts)-1], f.ports[len(f.ports)-1]
+}
+
+type fakeClusterAPI struct {
+	services  map[string]*corev1.Service
+	endpoints map[string]*corev1.Endpoints
+	pods      []*corev1.Pod
+}
+
+func (f *fakeClusterAPI) GetService(_ context.Context, namespace, name string) (*corev1.Service, error) {
+	svc, ok := f.services[namespace+"/"+name]
+	if !ok {
+		return nil, fmt.Errorf("service %s/%s not found", namespace, name)
+	}
+	return svc, nil
+}
+
+func (f *fakeClusterAPI) GetEndpoints(_ context.Context, namespace, name string) (*corev1.Endpoints, error) {
+	ep, ok := f.endpoints[namespace+"/"+name]
+	if !ok {
+		return nil, fmt.Errorf("endpoints %s/%s not found", namespace, name)
+	}
+	return ep, nil
+}
+
+func (f *fakeClusterAPI) ListServices(_ context.Context) (*corev1.ServiceList, error) {
+	list := &corev1.ServiceList{}
+	for _, svc := range f.services {
+		list.Items = append(list.Items, *svc)
+	}
+	return list, nil
+}
+
+func (f *fakeClusterAPI) ListPodsByIP(_ context.Context, ip string) (*corev1.PodList, error) {
+	list := &corev1.PodList{}
+	for _, pod := range f.pods {
+		if pod.Status.PodIP == ip {
+			list.Items = append(list.Items, *pod)
+		}
+	}
+	return list, nil
+}
+
+func TestParseServiceHost(t *testing.T) {
+	tests := []struct {
+		host      string
+		defaultNS string
+		wantSvc   string
+		wantNS    string
+		wantOK    bool
+	}{
+		{host: "jupyter.utils.svc.cluster.local", defaultNS: "default", wantSvc: "jupyter", wantNS: "utils", wantOK: true},
+		{host: "jupyter.utils.svc", defaultNS: "default", wantSvc: "jupyter", wantNS: "utils", wantOK: true},
+		{host: "jupyter.utils", defaultNS: "default", wantSvc: "jupyter", wantNS: "utils", wantOK: true},
+		{host: "jupyter", defaultNS: "utils", wantSvc: "jupyter", wantNS: "utils", wantOK: true},
+		{host: "api.example.com", defaultNS: "utils", wantOK: false},
+	}
+	for _, tt := range tests {
+		t.Run(tt.host, func(t *testing.T) {
+			gotSvc, gotNS, gotOK := parseServiceHost(tt.host, tt.defaultNS)
+			if gotSvc != tt.wantSvc || gotNS != tt.wantNS || gotOK != tt.wantOK {
+				t.Fatalf("parseServiceHost(%q) = (%q, %q, %v), want (%q, %q, %v)", tt.host, gotSvc, gotNS, gotOK, tt.wantSvc, tt.wantNS, tt.wantOK)
+			}
+		})
+	}
+}
+
+func TestResolveServiceToPod(t *testing.T) {
+	api := &fakeClusterAPI{
+		services: map[string]*corev1.Service{
+			"utils/jupyter": {
+				ObjectMeta: metav1.ObjectMeta{Name: "jupyter", Namespace: "utils"},
+				Spec: corev1.ServiceSpec{
+					ClusterIP: "10.103.245.38",
+					Ports:     []corev1.ServicePort{{Name: "http", Port: 8086}},
+				},
+			},
+		},
+		endpoints: map[string]*corev1.Endpoints{
+			"utils/jupyter": {
+				ObjectMeta: metav1.ObjectMeta{Name: "jupyter", Namespace: "utils"},
+				Subsets: []corev1.EndpointSubset{{
+					Addresses: []corev1.EndpointAddress{{
+						IP: "10.0.1.243",
+						TargetRef: &corev1.ObjectReference{
+							Kind:      "Pod",
+							Name:      "jupyter-0",
+							Namespace: "utils",
+						},
+					}},
+					Ports: []corev1.EndpointPort{{Name: "http", Port: 8086}},
+				}},
+			},
+		},
+	}
+
+	connector := &ClusterConnector{
+		Client:           api,
+		DefaultNamespace: "default",
+	}
+
+	target, err := connector.resolveServiceHost(context.Background(), "jupyter.utils.svc.cluster.local", 8086)
+	if err != nil {
+		t.Fatal(err)
+	}
+	if target.PodName != "jupyter-0" || target.Namespace != "utils" || target.PodPort != 8086 {
+		t.Fatalf("unexpected target: %#v", target)
+	}
+}
+
+func TestResolveTargetByClusterIP(t *testing.T) {
+	api := &fakeClusterAPI{
+		services: map[string]*corev1.Service{
+			"utils/jupyter": {
+				ObjectMeta: metav1.ObjectMeta{Name: "jupyter", Namespace: "utils"},
+				Spec: corev1.ServiceSpec{
+					ClusterIP: "10.103.245.38",
+					Ports: []corev1.ServicePort{{
+						Name:       "http",
+						Port:       8086,
+						TargetPort: intstr.FromInt(18086),
+					}},
+				},
+			},
+		},
+		endpoints: map[string]*corev1.Endpoints{
+			"utils/jupyter": {
+				ObjectMeta: metav1.ObjectMeta{Name: "jupyter", Namespace: "utils"},
+				Subsets: []corev1.EndpointSubset{{
+					Addresses: []corev1.EndpointAddress{{
+						IP: "10.0.1.243",
+						TargetRef: &corev1.ObjectReference{
+							Kind:      "Pod",
+							Name:      "jupyter-0",
+							Namespace: "utils",
+						},
+					}},
+					Ports: []corev1.EndpointPort{{Name: "http", Port: 18086}},
+				}},
+			},
+		},
+	}
+
+	connector := &ClusterConnector{Client: api, DefaultNamespace: "default"}
+	target, err := connector.resolveTarget(context.Background(), "10.103.245.38", 8086)
+	if err != nil {
+		t.Fatal(err)
+	}
+	if target.PodName != "jupyter-0" || target.Namespace != "utils" || target.PodPort != 18086 {
+		t.Fatalf("unexpected clusterIP target: %#v", target)
+	}
+}
+
+func TestResolveTargetByPodIP(t *testing.T) {
+	api := &fakeClusterAPI{
+		pods: []*corev1.Pod{{
+			ObjectMeta: metav1.ObjectMeta{Name: "jupyter-0", Namespace: "utils"},
+			Status:     corev1.PodStatus{Phase: corev1.PodRunning, PodIP: "10.0.1.243"},
+		}},
+	}
+
+	connector := &ClusterConnector{Client: api, DefaultNamespace: "default"}
+	target, err := connector.resolveTarget(context.Background(), "10.0.1.243", 8086)
+	if err != nil {
+		t.Fatal(err)
+	}
+	if target.PodName != "jupyter-0" || target.Namespace != "utils" || target.PodPort != 8086 {
+		t.Fatalf("unexpected podIP target: %#v", target)
+	}
+}
+
+func TestResolveTargetUnsupportedHost(t *testing.T) {
+	connector := &ClusterConnector{Client: &fakeClusterAPI{}, DefaultNamespace: "default"}
+	if _, err := connector.resolveTarget(context.Background(), "api.example.com", 443); err == nil {
+		t.Fatal("expected unsupported host resolution to fail")
+	}
+}
+
+func TestResolveServiceToPodNoReadyEndpoints(t *testing.T) {
+	api := &fakeClusterAPI{
+		services: map[string]*corev1.Service{
+			"utils/jupyter": {
+				ObjectMeta: metav1.ObjectMeta{Name: "jupyter", Namespace: "utils"},
+				Spec: corev1.ServiceSpec{
+					ClusterIP: "10.103.245.38",
+					Ports:     []corev1.ServicePort{{Name: "http", Port: 8086}},
+				},
+			},
+		},
+		endpoints: map[string]*corev1.Endpoints{
+			"utils/jupyter": {
+				ObjectMeta: metav1.ObjectMeta{Name: "jupyter", Namespace: "utils"},
+			},
+		},
+	}
+
+	connector := &ClusterConnector{Client: api, DefaultNamespace: "default"}
+	if _, err := connector.resolveServiceHost(context.Background(), "jupyter.utils.svc.cluster.local", 8086); err == nil {
+		t.Fatal("expected resolution to fail when service has no ready endpoints")
+	}
+}
+
+func TestSOCKS5ProxyPreservesHostname(t *testing.T) {
+	ctx, cancel := context.WithCancel(context.Background())
+	defer cancel()
+
+	backend := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		if r.URL.Path != "/tree/notebooks" {
+			http.NotFound(w, r)
+			return
+		}
+		_, _ = io.WriteString(w, "ok")
+	}))
+	defer backend.Close()
+
+	connector := &fakeConnector{addr: strings.TrimPrefix(backend.URL, "http://")}
+	ln, err := net.Listen("tcp", "127.0.0.1:0")
+	if err != nil {
+		t.Fatal(err)
+	}
+	defer ln.Close()
+
+	go func() {
+		_ = ServeSOCKS5(ctx, ln, connector)
+	}()
+
+	dialer, err := xproxy.SOCKS5("tcp", ln.Addr().String(), nil, xproxy.Direct)
+	if err != nil {
+		t.Fatal(err)
+	}
+	httpClient := &http.Client{
+		Timeout: 5 * time.Second,
+		Transport: &http.Transport{
+			DialContext: func(ctx context.Context, network, addr string) (net.Conn, error) {
+				return dialer.Dial(network, addr)
+			},
+		},
+	}
+
+	resp, err := httpClient.Get("http://jupyter.utils:8086/tree/notebooks")
+	if err != nil {
+		t.Fatal(err)
+	}
+	defer resp.Body.Close()
+	body, err := io.ReadAll(resp.Body)
+	if err != nil {
+		t.Fatal(err)
+	}
+	if string(body) != "ok" {
+		t.Fatalf("unexpected body %q", string(body))
+	}
+	host, port := connector.lastTarget()
+	if host != "jupyter.utils" || port != 8086 {
+		t.Fatalf("proxy connected to %s:%d, want jupyter.utils:8086", host, port)
+	}
+}
+
+func TestHTTPConnectProxyPreservesHostname(t *testing.T) {
+	ctx, cancel := context.WithCancel(context.Background())
+	defer cancel()
+
+	backend := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		if r.URL.Path != "/tree/notebooks" {
+			http.NotFound(w, r)
+			return
+		}
+		_, _ = io.WriteString(w, "ok")
+	}))
+	defer backend.Close()
+
+	connector := &fakeConnector{addr: strings.TrimPrefix(backend.URL, "http://")}
+	ln, err := net.Listen("tcp", "127.0.0.1:0")
+	if err != nil {
+		t.Fatal(err)
+	}
+	defer ln.Close()
+
+	go func() {
+		_ = ServeHTTPConnect(ctx, ln, connector)
+	}()
+
+	conn, err := net.Dial("tcp", ln.Addr().String())
+	if err != nil {
+		t.Fatal(err)
+	}
+	defer conn.Close()
+
+	_, err = fmt.Fprintf(conn, "CONNECT jupyter.utils:8086 HTTP/1.1\r\nHost: jupyter.utils:8086\r\n\r\n")
+	if err != nil {
+		t.Fatal(err)
+	}
+
+	reader := bufio.NewReader(conn)
+	resp, err := http.ReadResponse(reader, nil)
+	if err != nil {
+		t.Fatal(err)
+	}
+	if resp.StatusCode != http.StatusOK {
+		t.Fatalf("unexpected connect status %s", resp.Status)
+	}
+
+	_, err = fmt.Fprintf(conn, "GET /tree/notebooks HTTP/1.1\r\nHost: jupyter.utils:8086\r\nConnection: close\r\n\r\n")
+	if err != nil {
+		t.Fatal(err)
+	}
+	httpResp, err := http.ReadResponse(reader, nil)
+	if err != nil {
+		t.Fatal(err)
+	}
+	defer httpResp.Body.Close()
+	body, err := io.ReadAll(httpResp.Body)
+	if err != nil {
+		t.Fatal(err)
+	}
+	if string(body) != "ok" {
+		t.Fatalf("unexpected body %q", string(body))
+	}
+	host, port := connector.lastTarget()
+	if host != "jupyter.utils" || port != 8086 {
+		t.Fatalf("proxy connected to %s:%d, want jupyter.utils:8086", host, port)
+	}
+}
+
+func TestProxyOutE2ECase(t *testing.T) {
+	if testing.Short() {
+		t.Skip("skipping e2e in short mode")
+	}
+	if strings.TrimSpace(getenv("KUBEVPN_PROXY_OUT_E2E", "")) == "" {
+		t.Skip("set KUBEVPN_PROXY_OUT_E2E=1 to run real cluster proxy-out test")
+	}
+
+	host := getenv("KUBEVPN_PROXY_OUT_E2E_HOST", "jupyter.utils")
+	path := getenv("KUBEVPN_PROXY_OUT_E2E_PATH", "/tree/notebooks")
+	port := getenv("KUBEVPN_PROXY_OUT_E2E_PORT", "8086")
+
+	config, err := restConfigFromDefault()
+	if err != nil {
+		t.Fatal(err)
+	}
+	clusterAPI, clientset, err := NewClusterAPI(config)
+	if err != nil {
+		t.Fatal(err)
+	}
+
+	ctx, cancel := context.WithCancel(context.Background())
+	defer cancel()
+
+	connector := &ClusterConnector{
+		Client:           clusterAPI,
+		Forwarder:        NewPodDialer(config, clientset),
+		RESTConfig:       config,
+		DefaultNamespace: "default",
+	}
+	ln, err := net.Listen("tcp", "127.0.0.1:0")
+	if err != nil {
+		t.Fatal(err)
+	}
+	defer ln.Close()
+	go func() {
+		_ = ServeSOCKS5(ctx, ln, connector)
+	}()
+
+	dialer, err := xproxy.SOCKS5("tcp", ln.Addr().String(), nil, xproxy.Direct)
+	if err != nil {
+		t.Fatal(err)
+	}
+	httpClient := &http.Client{
+		Timeout: 15 * time.Second,
+		Transport: &http.Transport{
+			DialContext: func(ctx context.Context, network, addr string) (net.Conn, error) {
+				return dialer.Dial(network, addr)
+			},
+		},
+		CheckRedirect: func(req *http.Request, via []*http.Request) error {
+			return http.ErrUseLastResponse
+		},
+	}
+
+	targetURL := (&url.URL{Scheme: "http", Host: net.JoinHostPort(host, port), Path: path}).String()
+	resp, err := httpClient.Get(targetURL)
+	if err != nil {
+		t.Fatal(err)
+	}
+	defer resp.Body.Close()
+	if resp.StatusCode != http.StatusOK && resp.StatusCode != http.StatusFound && resp.StatusCode != http.StatusUnauthorized {
+		body, _ := io.ReadAll(resp.Body)
+		t.Fatalf("unexpected status %s body=%s", resp.Status, string(body))
+	}
+}
diff --git a/pkg/localproxy/socks5.go b/pkg/localproxy/socks5.go
new file mode 100644
index 00000000..192c6fa7
--- /dev/null
+++ b/pkg/localproxy/socks5.go
@@ -0,0 +1,132 @@
+package localproxy
+
+import (
+	"bufio"
+	"context"
+	"encoding/binary"
+	"fmt"
+	"io"
+	"net"
+)
+
+const (
+	socks5Version       = 5
+	socks5AuthNone      = 0
+	socks5CmdConnect    = 1
+	socks5AddrTypeIPv4  = 1
+	socks5AddrTypeFQDN  = 3
+	socks5AddrTypeIPv6  = 4
+	socks5ReplySuccess  = 0
+	socks5ReplyGeneral  = 1
+	socks5ReplyNotAllow = 2
+	socks5ReplyUnsup    = 7
+)
+
+func ServeSOCKS5(ctx context.Context, ln net.Listener, connector Connector) error {
+	for {
+		conn, err := ln.Accept()
+		if err != nil {
+			select {
+			case <-ctx.Done():
+				return nil
+			default:
+				return err
+			}
+		}
+		go handleSOCKS5Conn(ctx, conn, connector)
+	}
+}
+
+func handleSOCKS5Conn(ctx context.Context, conn net.Conn, connector Connector) {
+	defer conn.Close()
+
+	reader := bufio.NewReader(conn)
+	if err := socks5Handshake(reader, conn); err != nil {
+		return
+	}
+	host, port, err := socks5ReadRequest(reader)
+	if err != nil {
+		_ = writeSOCKS5Reply(conn, socks5ReplyGeneral)
+		return
+	}
+
+	remote, err := connector.Connect(ctx, host, port)
+	if err != nil {
+		_ = writeSOCKS5Reply(conn, socks5ReplyGeneral)
+		return
+	}
+	defer remote.Close()
+
+	if err := writeSOCKS5Reply(conn, socks5ReplySuccess); err != nil {
+		return
+	}
+	relayConns(conn, remote)
+}
+
+func socks5Handshake(reader *bufio.Reader, conn net.Conn) error {
+	header := make([]byte, 2)
+	if _, err := io.ReadFull(reader, header); err != nil {
+		return err
+	}
+	if header[0] != socks5Version {
+		return fmt.Errorf("unsupported socks version %d", header[0])
+	}
+	methods := make([]byte, int(header[1]))
+	if _, err := io.ReadFull(reader, methods); err != nil {
+		return err
+	}
+	_, err := conn.Write([]byte{socks5Version, socks5AuthNone})
+	return err
+}
+
+func socks5ReadRequest(reader *bufio.Reader) (string, int, error) {
+	header := make([]byte, 4)
+	if _, err := io.ReadFull(reader, header); err != nil {
+		return "", 0, err
+	}
+	if header[0] != socks5Version {
+		return "", 0, fmt.Errorf("unsupported socks request version %d", header[0])
+	}
+	if header[1] != socks5CmdConnect {
+		return "", 0, fmt.Errorf("unsupported socks command %d", header[1])
+	}
+
+	var host string
+	switch header[3] {
+	case socks5AddrTypeIPv4:
+		addr := make([]byte, 4)
+		if _, err := io.ReadFull(reader, addr); err != nil {
+			return "", 0, err
+		}
+		host = net.IP(addr).String()
+	case socks5AddrTypeIPv6:
+		addr := make([]byte, 16)
+		if _, err := io.ReadFull(reader, addr); err != nil {
+			return "", 0, err
+		}
+		host = net.IP(addr).String()
+	case socks5AddrTypeFQDN:
+		length, err := reader.ReadByte()
+		if err != nil {
+			return "", 0, err
+		}
+		addr := make([]byte, int(length))
+		if _, err := io.ReadFull(reader, addr); err != nil {
+			return "", 0, err
+		}
+		host = string(addr)
+	default:
+		return "", 0, fmt.Errorf("unsupported socks address type %d", header[3])
+	}
+
+	portBuf := make([]byte, 2)
+	if _, err := io.ReadFull(reader, portBuf); err != nil {
+		return "", 0, err
+	}
+	return host, int(binary.BigEndian.Uint16(portBuf)), nil
+}
+
+func writeSOCKS5Reply(conn net.Conn, rep byte) error {
+	_, err := conn.Write([]byte{socks5Version, rep, 0, socks5AddrTypeIPv4, 0, 0, 0, 0, 0, 0})
+	return err
+}
diff --git a/pkg/localproxy/test_helpers_test.go b/pkg/localproxy/test_helpers_test.go
new file mode 100644
index 00000000..3e9f76ea
--- /dev/null
+++ b/pkg/localproxy/test_helpers_test.go
@@ -0,0 +1,21 @@
+package localproxy
+
+import (
+	"os"
+
+	"k8s.io/client-go/rest"
+	"k8s.io/client-go/tools/clientcmd"
+)
+
+func getenv(key, fallback string) string {
+	if v := os.Getenv(key); v != "" {
+		return v
+	}
+	return fallback
+}
+
+func restConfigFromDefault() (*rest.Config, error) {
+	loadingRules := clientcmd.NewDefaultClientConfigLoadingRules()
+	configOverrides := &clientcmd.ConfigOverrides{}
+	return clientcmd.NewNonInteractiveDeferredLoadingClientConfig(loadingRules, configOverrides).ClientConfig()
+}
diff --git a/pkg/util/pod.go b/pkg/util/pod.go
index 1843c69f..c94e3bf4 100644
--- a/pkg/util/pod.go
+++ b/pkg/util/pod.go
@@ -10,6 +10,7 @@ import (
 	"os"
 	"strconv"
 	"strings"
+	"sync"
 	"text/tabwriter"
 	"time"
 
@@ -19,6 +20,7 @@ import (
 	"k8s.io/apimachinery/pkg/apis/meta/v1"
 	"k8s.io/apimachinery/pkg/fields"
 	"k8s.io/apimachinery/pkg/util/httpstream"
+	utilruntime "k8s.io/apimachinery/pkg/util/runtime"
 	"k8s.io/apimachinery/pkg/util/sets"
 	"k8s.io/apimachinery/pkg/watch"
 	"k8s.io/cli-runtime/pkg/genericiooptions"
@@ -145,6 +147,7 @@ func PortForwardPod(config *rest.Config, clientset *rest.RESTClient, podName, na
 		return err
 	}
 
+	suppressExpectedPortForwardCloseErrors()
 	defer forwarder.Close()
 
 	var errChan = make(chan error, 1)
@@ -160,6 +163,25 @@ func PortForwardPod(config *rest.Config, clientset *rest.RESTClient, podName, na
 	}
 }
 
+var portForwardErrorHandlerOnce sync.Once
+
+func suppressExpectedPortForwardCloseErrors() {
+	portForwardErrorHandlerOnce.Do(func() {
+		prev := append([]utilruntime.ErrorHandler(nil), utilruntime.ErrorHandlers...)
+		utilruntime.ErrorHandlers = []utilruntime.ErrorHandler{
+			func(ctx context.Context, err error, msg string, keysAndValues ...interface{}) {
+				if err != nil && strings.Contains(strings.ToLower(err.Error()), "error closing listener: close tcp") &&
+					strings.Contains(strings.ToLower(err.Error()), "use of closed network connection") {
+					return
+				}
+				for _, handler := range prev {
+					handler(ctx, err, msg, keysAndValues...)
+				}
+			},
+		}
+	})
+}
+
 func Shell(ctx context.Context, clientset *kubernetes.Clientset, config *rest.Config, podName, containerName, ns string, cmd []string) (string, error) {
 	stdin, _, _ := term.StdStreams()
 	buf := bytes.NewBuffer(nil)