Reduce MCP discovery timeouts

When MCP plugins are unreachable, the discovery times out too slowly and blocks commands. Reduced the overall discovery timeout from 10s to 4s and per-server timeout from 5s to 2s. Made the per-server timeout configurable.

.github/badges/coverage.svg

@@ -1 +1 @@
-<svg xmlns="http://www.w3.org/2000/svg" width="108" height="20" role="img" aria-label="coverage: 48.8%"><title>coverage: 48.8%</title><linearGradient id="s" x2="0" y2="100%"><stop offset="0" stop-color="#bbb" stop-opacity=".1"/><stop offset="1" stop-opacity=".1"/></linearGradient><clipPath id="r"><rect width="108" height="20" rx="3" fill="#fff"/></clipPath><g clip-path="url(#r)"><rect width="61" height="20" fill="#555"/><rect x="61" width="47" height="20" fill="#e05d44"/><rect width="108" height="20" fill="url(#s)"/></g><g fill="#fff" text-anchor="middle" font-family="Verdana,Geneva,DejaVu Sans,sans-serif" text-rendering="geometricPrecision" font-size="110"><text aria-hidden="true" x="315" y="150" fill="#010101" fill-opacity=".3" transform="scale(.1)" textLength="510">coverage</text><text x="315" y="140" transform="scale(.1)" fill="#fff" textLength="510">coverage</text><text aria-hidden="true" x="835" y="150" fill="#010101" fill-opacity=".3" transform="scale(.1)" textLength="370">48.8%</text><text x="835" y="140" transform="scale(.1)" fill="#fff" textLength="370">48.8%</text></g></svg>
+<svg xmlns="http://www.w3.org/2000/svg" width="108" height="20" role="img" aria-label="coverage: 48.7%"><title>coverage: 48.7%</title><linearGradient id="s" x2="0" y2="100%"><stop offset="0" stop-color="#bbb" stop-opacity=".1"/><stop offset="1" stop-opacity=".1"/></linearGradient><clipPath id="r"><rect width="108" height="20" rx="3" fill="#fff"/></clipPath><g clip-path="url(#r)"><rect width="61" height="20" fill="#555"/><rect x="61" width="47" height="20" fill="#e05d44"/><rect width="108" height="20" fill="url(#s)"/></g><g fill="#fff" text-anchor="middle" font-family="Verdana,Geneva,DejaVu Sans,sans-serif" text-rendering="geometricPrecision" font-size="110"><text aria-hidden="true" x="315" y="150" fill="#010101" fill-opacity=".3" transform="scale(.1)" textLength="510">coverage</text><text x="315" y="140" transform="scale(.1)" fill="#fff" textLength="510">coverage</text><text aria-hidden="true" x="835" y="150" fill="#010101" fill-opacity=".3" transform="scale(.1)" textLength="370">48.7%</text><text x="835" y="140" transform="scale(.1)" fill="#fff" textLength="370">48.7%</text></g></svg>

internal/app/plugin_cmd.go

@@ -190,6 +190,8 @@ func newPluginRemoveCommand() *cobra.Command {
 		Args:              cobra.ExactArgs(1),
 		DisableAutoGenTag: true,
 		RunE: func(cmd *cobra.Command, args []string) error {
+			// Stop stdio clients before removing to release file locks
+			StopStdioClientsByPlugin(args[0])
 			keepData, _ := cmd.Flags().GetBool("keep-data")
 			loader := plugin.NewLoader(RawVersion())
 			if err := loader.RemovePlugin(args[0], keepData); err != nil {

internal/app/root.go

@@ -57,9 +57,17 @@ type outputFileContextKey struct{}
 const recoveryEventStderrPrefix = "RECOVERY_EVENT_ID="
 
 // Execute runs the root command and returns the process exit code.
-func Execute() int {
+func Execute() (exitCode int) {
+	defer func() {
+		if r := recover(); r != nil {
+			fmt.Fprintf(os.Stderr, "Error: internal panic: %v\n", r)
+			exitCode = 5
+		}
+	}()
+
 	timing := NewTimingCollector()
 	defer func() {
+		StopAllStdioClients() // Ensure child processes are terminated on exit
 		timing.PrintIfEnabled()
 		timing.WriteReportIfEnabled(RawVersion(), SanitizeCommand(os.Args))
 	}()
@@ -265,6 +273,7 @@ func NewRootCommandWithEngine(rootCtx context.Context, engine *pipeline.Engine)
 			return nil
 		},
 		PersistentPostRunE: func(cmd *cobra.Command, args []string) error {
+			StopAllStdioClients()
 			CloseFileLogger()
 			return closeOutputSink(cmd)
 		},
@@ -1029,8 +1038,23 @@ func loadPlugins(engine *pipeline.Engine, runner executor.Runner) []*cobra.Comma
 
 	// 0a. Ensure default managed plugins are installed (first-run bootstrap).
 	updater := plugin.NewUpdater(pluginLoader.PluginsDir, RawVersion())
-	accessToken, tokenErr := loadSkillAccessToken()
-	if tokenErr == nil && accessToken != "" {
+	// Load TokenData once; reuse for plugin bootstrap, updates, and stdio injection.
+	tokenData, _ := authpkg.LoadTokenData(defaultConfigDir())
+	var userCtx *plugin.UserContext
+	if tokenData != nil {
+		// Inject user context if either UserID or CorpID is present.
+		if tokenData.UserID != "" || tokenData.CorpID != "" {
+			userCtx = &plugin.UserContext{
+				UserID: tokenData.UserID,
+				CorpID: tokenData.CorpID,
+			}
+		}
+	}
+	accessToken := ""
+	if tokenData != nil && tokenData.IsAccessTokenValid() {
+		accessToken = tokenData.AccessToken
+	}
+	if accessToken != "" {
 		bootstrapCtx, bootstrapCancel := context.WithTimeout(context.Background(), 2*time.Minute)
 		installed := updater.EnsureManaged(bootstrapCtx, accessToken, os.Stderr)
 		bootstrapCancel()
@@ -1114,7 +1138,7 @@ func loadPlugins(engine *pipeline.Engine, runner executor.Runner) []*cobra.Comma
 
 	// 4. Start stdio MCP servers, discover tools, and build CLI commands
 	for _, p := range allPlugins {
-		for _, sc := range p.StdioClients() {
+		for _, sc := range p.StdioClients(userCtx) {
 			// Use background context so the subprocess lives for the CLI
 			// process lifetime (not killed by a short timeout).
 			if err := sc.Client.Start(context.Background()); err != nil {
@@ -1405,7 +1429,8 @@ func registerStdioServer(p *plugin.Plugin, sc plugin.StdioServerClient, runner e
 	}
 
 	AppendDynamicServer(descriptor)
-	RegisterStdioClient(serverID, sc.Client)
+	// Register with pluginName/serverKey format for cleanup by plugin name
+	RegisterStdioClient(p.Manifest.Name+"/"+serverID, sc.Client)
 
 	// Convert tool descriptors to DetailTool entries for flag generation.
 	detailsByID := make(map[string][]market.DetailTool)

internal/app/root_execute_test.go

@@ -408,3 +408,33 @@ func TestPrintExecutionError_RawStderrError_takes_precedence_over_JSON_mode(t *t
 		t.Fatalf("stderr = %q, want raw PAT JSON", stderr.String())
 	}
 }
+
+// simulateExecuteWithPanic mirrors the recovery pattern in Execute():
+// named return + defer recover → exitCode = 5 on panic.
+func simulateExecuteWithPanic(doPanic bool) (exitCode int) {
+	defer func() {
+		if r := recover(); r != nil {
+			exitCode = 5
+		}
+	}()
+	if doPanic {
+		panic("test panic")
+	}
+	return 0
+}
+
+func TestExecute_panic_recovery_returns_exit_5(t *testing.T) {
+	t.Parallel()
+	code := simulateExecuteWithPanic(true)
+	if code != 5 {
+		t.Fatalf("panic recovery exitCode = %d, want 5", code)
+	}
+}
+
+func TestExecute_no_panic_returns_0(t *testing.T) {
+	t.Parallel()
+	code := simulateExecuteWithPanic(false)
+	if code != 0 {
+		t.Fatalf("no-panic exitCode = %d, want 0", code)
+	}
+}

internal/app/skill_command.go

@@ -75,7 +75,7 @@ type findSkillsResponse struct {
 	Result    []CliSkillDTO `json:"result,omitempty"`
 }
 
-// CliSkillDTO mirrors the old cli response payload for `skill find`.
+// CliSkillDTO mirrors the old cli response payload for `skill search`.
 type CliSkillDTO struct {
 	SkillID string `json:"skillId"`
 	Name    string `json:"name"`
@@ -117,10 +117,11 @@ func buildSkillCommand() *cobra.Command {
 	}
 
 	cmd.AddCommand(
-		newSkillAddCommand(),
+		newSkillInstallCommand(),
 		newSkillGetCommand(),
-		newSkillFindCommand(),
-		newSkillSearchHintCommand(),
+		newSkillSearchCommand(),
+		newSkillFindHintCommand(),
+		newSkillAddHintCommand(),
 	)
 	return cmd
 }
@@ -139,36 +140,37 @@ func newSkillGetCommand() *cobra.Command {
 	return cmd
 }
 
-func newSkillFindCommand() *cobra.Command {
+func newSkillSearchCommand() *cobra.Command {
 	cmd := &cobra.Command{
-		Use:               "find",
+		Use:               "search",
 		Short:             "从钉钉技能市场搜索技能",
 		Long:              "从钉钉技能市场搜索技能，根据关键词返回匹配的技能列表。",
-		Example:           "  dws skill find --context 关键词",
+		Example:           "  dws skill search --query 关键词",
 		DisableAutoGenTag: true,
 		RunE:              runSkillFind,
 	}
-	cmd.Flags().String("context", "", "搜索关键词（必填）")
-	_ = cmd.MarkFlagRequired("context")
+	cmd.Flags().String("query", "", "搜索关键词（必填）")
+	_ = cmd.MarkFlagRequired("query")
+	cmd.Flags().String("scopes", "", "查询范围，空格分隔。备选值：DingtalkMarket（钉钉市场）、OrgInternal（企业内部）。为空默认查市场技能")
 	return cmd
 }
 
-func newSkillSearchHintCommand() *cobra.Command {
+func newSkillFindHintCommand() *cobra.Command {
 	return &cobra.Command{
-		Use:               "search",
-		Short:             "兼容旧用法，提示使用 skill find",
+		Use:               "find",
+		Short:             "兼容旧用法，提示使用 skill search",
 		Hidden:            true,
 		DisableAutoGenTag: true,
 		RunE: func(cmd *cobra.Command, args []string) error {
-			_, _ = fmt.Fprintln(cmd.OutOrStdout(), "use: dws skill find --context <关键词>")
+			_, _ = fmt.Fprintln(cmd.OutOrStdout(), "use: dws skill search --query <关键词>")
 			return nil
 		},
 	}
 }
 
-func newSkillAddCommand() *cobra.Command {
+func newSkillInstallCommand() *cobra.Command {
 	cmd := &cobra.Command{
-		Use:   "add <skillId> <target>",
+		Use:   "install <skillId> <target>",
 		Short: "下载并安装技能到指定目录",
 		Long: fmt.Sprintf(`从钉钉技能市场下载技能并安装到指定 Agent 目录。
 
@@ -185,9 +187,9 @@ func newSkillAddCommand() *cobra.Command {
   .        -> 当前目录
 
 示例:
-  dws skill add skill-123 qoder     # 安装到 ~/.qoder/skills/
-  dws skill add skill-123 claude    # 安装到 ~/.claude/skills/
-  dws skill add skill-123 .         # 安装到当前目录`, supportedTargets()),
+  dws skill install skill-123 qoder     # 安装到 ~/.qoder/skills/
+  dws skill install skill-123 claude    # 安装到 ~/.claude/skills/
+  dws skill install skill-123 .         # 安装到当前目录`, supportedTargets()),
 		Args:              cobra.ExactArgs(2),
 		DisableAutoGenTag: true,
 		RunE:              runSkillAdd,
@@ -196,6 +198,19 @@ func newSkillAddCommand() *cobra.Command {
 	return cmd
 }
 
+func newSkillAddHintCommand() *cobra.Command {
+	return &cobra.Command{
+		Use:               "add",
+		Short:             "兼容旧用法，提示使用 skill install",
+		Hidden:            true,
+		DisableAutoGenTag: true,
+		RunE: func(cmd *cobra.Command, args []string) error {
+			_, _ = fmt.Fprintln(cmd.OutOrStdout(), "use: dws skill install <skillId> <target>")
+			return nil
+		},
+	}
+}
+
 func runSkillGet(cmd *cobra.Command, args []string) error {
 	skillID, _ := cmd.Flags().GetString("skill-id")
 	accessToken, err := loadSkillAccessToken()
@@ -216,13 +231,17 @@ func runSkillGet(cmd *cobra.Command, args []string) error {
 }
 
 func runSkillFind(cmd *cobra.Command, args []string) error {
-	keyword, _ := cmd.Flags().GetString("context")
+	keyword, _ := cmd.Flags().GetString("query")
+	scopes, _ := cmd.Flags().GetString("scopes")
 	accessToken, err := loadSkillAccessToken()
 	if err != nil {
 		return err
 	}
 
 	apiURL := fmt.Sprintf("%s/cli/find-skills?keyword=%s", skillAPIHost(), url.QueryEscape(strings.TrimSpace(keyword)))
+	if scopes != "" {
+		apiURL += "&scopes=" + url.QueryEscape(scopes)
+	}
 	req, err := http.NewRequestWithContext(cmd.Context(), http.MethodGet, apiURL, nil)
 	if err != nil {
 		return apperrors.NewInternal(fmt.Sprintf("failed to create request: %v", err))

internal/app/skill_command_test.go

@@ -319,7 +319,7 @@ func TestExtractSkillZipPreventZipSlip(t *testing.T) {
 	}
 }
 
-func TestSkillAddCommandValidation(t *testing.T) {
+func TestSkillInstallCommandValidation(t *testing.T) {
 	tests := []struct {
 		name    string
 		args    []string
@@ -328,19 +328,19 @@ func TestSkillAddCommandValidation(t *testing.T) {
 	}{
 		{
 			name:    "missing arguments",
-			args:    []string{"skill", "add"},
+			args:    []string{"skill", "install"},
 			wantErr: true,
 			errMsg:  "accepts 2 arg(s)",
 		},
 		{
 			name:    "missing target",
-			args:    []string{"skill", "add", "skill-123"},
+			args:    []string{"skill", "install", "skill-123"},
 			wantErr: true,
 			errMsg:  "accepts 2 arg(s)",
 		},
 		{
 			name:    "too many arguments",
-			args:    []string{"skill", "add", "skill-123", "qoder", "extra"},
+			args:    []string{"skill", "install", "skill-123", "qoder", "extra"},
 			wantErr: true,
 			errMsg:  "accepts 2 arg(s)",
 		},
@@ -366,7 +366,7 @@ func TestSkillAddCommandValidation(t *testing.T) {
 	}
 }
 
-func TestSkillAddInvalidTarget(t *testing.T) {
+func TestSkillInstallInvalidTarget(t *testing.T) {
 	// Setup: Create config directory with valid token
 	tempDir := t.TempDir()
 	configDir := filepath.Join(tempDir, "config")
@@ -384,7 +384,7 @@ func TestSkillAddInvalidTarget(t *testing.T) {
 	}
 
 	cmd := NewRootCommand()
-	cmd.SetArgs([]string{"skill", "add", "skill-123", "invalid-target"})
+	cmd.SetArgs([]string{"skill", "install", "skill-123", "invalid-target"})
 
 	var out bytes.Buffer
 	cmd.SetOut(&out)
@@ -399,7 +399,7 @@ func TestSkillAddInvalidTarget(t *testing.T) {
 	}
 }
 
-func TestSkillAddRequiresAuth(t *testing.T) {
+func TestSkillInstallRequiresAuth(t *testing.T) {
 	// Setup: Create config directory without token
 	tempDir := t.TempDir()
 	configDir := filepath.Join(tempDir, "config")
@@ -411,7 +411,7 @@ func TestSkillAddRequiresAuth(t *testing.T) {
 	}
 
 	cmd := NewRootCommand()
-	cmd.SetArgs([]string{"skill", "add", "skill-123", "qoder"})
+	cmd.SetArgs([]string{"skill", "install", "skill-123", "qoder"})
 
 	var out bytes.Buffer
 	cmd.SetOut(&out)
@@ -563,16 +563,16 @@ func TestSkillCommandHelp(t *testing.T) {
 	if !strings.Contains(output, "技能") {
 		t.Errorf("help should mention '技能', got: %s", output)
 	}
-	for _, subcmd := range []string{"add", "find", "get"} {
+	for _, subcmd := range []string{"install", "search", "get"} {
 		if !strings.Contains(output, subcmd) {
 			t.Errorf("help should mention %q subcommand, got: %s", subcmd, output)
 		}
 	}
 }
 
-func TestSkillAddCommandHelp(t *testing.T) {
+func TestSkillInstallCommandHelp(t *testing.T) {
 	cmd := NewRootCommand()
-	cmd.SetArgs([]string{"skill", "add", "--help"})
+	cmd.SetArgs([]string{"skill", "install", "--help"})
 
 	var out bytes.Buffer
 	cmd.SetOut(&out)
@@ -609,9 +609,9 @@ func TestSkillGetCommandValidation(t *testing.T) {
 	}
 }
 
-func TestSkillFindCommandValidation(t *testing.T) {
+func TestSkillSearchCommandValidation(t *testing.T) {
 	cmd := NewRootCommand()
-	cmd.SetArgs([]string{"skill", "find"})
+	cmd.SetArgs([]string{"skill", "search"})
 
 	var out bytes.Buffer
 	cmd.SetOut(&out)
@@ -626,9 +626,9 @@ func TestSkillFindCommandValidation(t *testing.T) {
 	}
 }
 
-func TestSkillSearchHintCommand(t *testing.T) {
+func TestSkillFindHintCommand(t *testing.T) {
 	cmd := NewRootCommand()
-	cmd.SetArgs([]string{"skill", "search"})
+	cmd.SetArgs([]string{"skill", "find"})
 
 	var out bytes.Buffer
 	cmd.SetOut(&out)
@@ -637,7 +637,7 @@ func TestSkillSearchHintCommand(t *testing.T) {
 	if err := cmd.Execute(); err != nil {
 		t.Fatalf("Execute() error = %v", err)
 	}
-	if !strings.Contains(out.String(), "dws skill find --context") {
+	if !strings.Contains(out.String(), "dws skill search --query") {
 		t.Fatalf("output = %q, want legacy hint", out.String())
 	}
 }

internal/app/stdio_registry.go

@@ -14,6 +14,7 @@
 package app
 
 import (
+	"log/slog"
 	"strings"
 	"sync"
 
@@ -37,11 +38,24 @@ func RegisterStdioClient(productID string, client *transport.StdioClient) {
 }
 
 // LookupStdioClient returns the StdioClient registered for the given product ID.
+// The productID can be either the full key (pluginName/serverKey) or just the serverKey.
+// This supports backward compatibility with existing CanonicalProduct values.
 func LookupStdioClient(productID string) (*transport.StdioClient, bool) {
 	stdioMu.RLock()
 	defer stdioMu.RUnlock()
-	c, ok := stdioClients[productID]
-	return c, ok
+	// Try exact match first
+	if c, ok := stdioClients[productID]; ok {
+		return c, true
+	}
+	// If not found, try matching by serverKey suffix (for backward compatibility)
+	for id, c := range stdioClients {
+		if idx := strings.LastIndex(id, "/"); idx >= 0 {
+			if id[idx+1:] == productID {
+				return c, true
+			}
+		}
+	}
+	return nil, false
 }
 
 // StdioEndpoint returns a virtual endpoint URL for a stdio-based MCP server.
@@ -54,3 +68,52 @@ func StdioEndpoint(pluginName, serverKey string) string {
 func IsStdioEndpoint(endpoint string) bool {
 	return strings.HasPrefix(endpoint, stdioEndpointScheme)
 }
+
+// StopAllStdioClients stops all registered stdio clients.
+// This should be called on program exit to terminate child processes.
+func StopAllStdioClients() {
+	stdioMu.Lock()
+	defer stdioMu.Unlock()
+	for id, client := range stdioClients {
+		if err := client.Stop(); err != nil {
+			slog.Warn("failed to stop stdio client", "id", id, "error", err)
+		}
+	}
+	stdioClients = make(map[string]*transport.StdioClient)
+}
+
+// StopStdioClient stops a specific stdio client by product ID.
+// Returns true if the client was found and stopped, false otherwise.
+func StopStdioClient(productID string) bool {
+	stdioMu.Lock()
+	defer stdioMu.Unlock()
+	client, ok := stdioClients[productID]
+	if !ok {
+		return false
+	}
+	if err := client.Stop(); err != nil {
+		slog.Warn("failed to stop stdio client", "id", productID, "error", err)
+	}
+	delete(stdioClients, productID)
+	return true
+}
+
+// StopStdioClientsByPlugin stops all stdio clients belonging to a plugin.
+// The productID format is "pluginName/serverKey". This function stops all
+// clients whose productID has the given pluginName prefix.
+func StopStdioClientsByPlugin(pluginName string) int {
+	stdioMu.Lock()
+	defer stdioMu.Unlock()
+	prefix := pluginName + "/"
+	count := 0
+	for id, client := range stdioClients {
+		if len(id) > len(prefix) && id[:len(prefix)] == prefix {
+			if err := client.Stop(); err != nil {
+				slog.Warn("failed to stop stdio client", "id", id, "error", err)
+			}
+			delete(stdioClients, id)
+			count++
+		}
+	}
+	return count
+}

internal/cli/loader.go

@@ -107,7 +107,7 @@ const (
 	DefaultMarketBaseURL = "https://mcp.dingtalk.com"
 
 	// defaultDiscoveryTimeout bounds the time spent on live registry discovery.
-	defaultDiscoveryTimeout = 10 * time.Second
+	defaultDiscoveryTimeout = 4 * time.Second
 )
 
 type CatalogLoader interface {

internal/discovery/service.go

@@ -53,12 +53,13 @@ const (
 var errCLIServerSkipped = errors.New("server marked cli.skip")
 
 type Service struct {
-	MarketClient *market.Client
-	Transport    *transport.Client
-	Cache        *cache.Store
-	Tenant       string
-	AuthIdentity string
-	Logger       *slog.Logger
+	MarketClient     *market.Client
+	Transport        *transport.Client
+	Cache            *cache.Store
+	Tenant           string
+	AuthIdentity     string
+	Logger           *slog.Logger
+	PerServerTimeout time.Duration // overrides perServerDiscoveryTimeout when > 0
 }
 
 type RuntimeServer struct {
@@ -170,7 +171,7 @@ func (s *Service) DiscoverServerRuntime(ctx context.Context, server market.Serve
 	}, nil
 }
 
-const perServerDiscoveryTimeout = 5 * time.Second
+const perServerDiscoveryTimeout = 2 * time.Second
 
 func (s *Service) DiscoverAllRuntime(ctx context.Context, servers []market.ServerDescriptor) ([]RuntimeServer, []RuntimeFailure) {
 	type discoveryResult struct {
@@ -189,12 +190,16 @@ func (s *Service) DiscoverAllRuntime(ctx context.Context, servers []market.Serve
 	}
 
 	ch := make(chan discoveryResult, len(filtered))
+	serverTimeout := s.PerServerTimeout
+	if serverTimeout <= 0 {
+		serverTimeout = perServerDiscoveryTimeout
+	}
 	var wg sync.WaitGroup
 	for _, srv := range filtered {
 		wg.Add(1)
 		go func(server market.ServerDescriptor) {
 			defer wg.Done()
-			serverCtx, cancel := context.WithTimeout(ctx, perServerDiscoveryTimeout)
+			serverCtx, cancel := context.WithTimeout(ctx, serverTimeout)
 			defer cancel()
 			start := time.Now()
 			rs, err := s.DiscoverServerRuntime(serverCtx, server)

internal/discovery/service_test.go

@@ -6,6 +6,7 @@ import (
 	"net/http"
 	"net/http/httptest"
 	"testing"
+	"time"
 
 	"github.com/DingTalk-Real-AI/dingtalk-workspace-cli/internal/cache"
 	"github.com/DingTalk-Real-AI/dingtalk-workspace-cli/internal/market"
@@ -435,6 +436,60 @@ func TestParseDetailSchema(t *testing.T) {
 	}
 }
 
+func TestDiscoverAllRuntime_TimeoutFallsBackToCache(t *testing.T) {
+	t.Parallel()
+
+	// done signals slow handlers to exit so srv.Close() can complete.
+	done := make(chan struct{})
+	// Server that blocks until signalled (simulates an unreachable MCP server).
+	srv := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		select {
+		case <-done:
+		case <-time.After(10 * time.Second):
+		}
+		http.Error(w, "timeout", http.StatusServiceUnavailable)
+	}))
+	// LIFO: close(done) runs first so handlers exit, then srv.Close() completes.
+	defer srv.Close()
+	defer close(done)
+
+	svc := newTestService(t, srv.URL, srv)
+	// Override timeout so the test completes quickly.
+	svc.PerServerTimeout = 80 * time.Millisecond
+
+	server := market.ServerDescriptor{
+		Key:      "slow-server",
+		Endpoint: srv.URL + "/mcp",
+	}
+
+	// Pre-populate the cache so the fallback has data to return.
+	partition := "test-tenant/test-identity"
+	_ = svc.Cache.SaveTools(partition, server.Key, cache.ToolsSnapshot{
+		ServerKey:       server.Key,
+		ProtocolVersion: "2025-03-26",
+		Tools: []transport.ToolDescriptor{
+			{Name: "cached-tool", Description: "from cache"},
+		},
+	})
+
+	results, failures := svc.DiscoverAllRuntime(context.Background(), []market.ServerDescriptor{server})
+	if len(failures) != 0 {
+		t.Fatalf("failures count = %d, want 0 (expected cache fallback)", len(failures))
+	}
+	if len(results) != 1 {
+		t.Fatalf("results count = %d, want 1", len(results))
+	}
+	if !results[0].Degraded {
+		t.Fatal("cache fallback result should be degraded")
+	}
+	if len(results[0].Tools) == 0 {
+		t.Fatal("expected cached tools to be returned")
+	}
+	if results[0].Tools[0].Name != "cached-tool" {
+		t.Fatalf("tool name = %q, want cached-tool", results[0].Tools[0].Name)
+	}
+}
+
 func TestPartition(t *testing.T) {
 	t.Parallel()
 	svc := &Service{Tenant: "corp1", AuthIdentity: "user1"}

internal/plugin/converter.go

@@ -24,6 +24,13 @@ import (
 	"github.com/DingTalk-Real-AI/dingtalk-workspace-cli/internal/transport"
 )
 
+// UserContext holds the minimal user identity fields injected into
+// stdio plugin subprocesses via environment variables.
+type UserContext struct {
+	UserID string
+	CorpID string
+}
+
 // StdioServerClient pairs a transport.StdioClient with its server key.
 type StdioServerClient struct {
 	Key    string
@@ -31,8 +38,11 @@ type StdioServerClient struct {
 }
 
 // StdioClients returns StdioClient instances for all stdio-type MCP
-// servers declared by this plugin.
-func (p *Plugin) StdioClients() []StdioServerClient {
+// servers declared by this plugin. uc is the current user's identity;
+// if non-nil, DWS_USER_ID and DWS_CORP_ID are injected as environment
+// variables so that the subprocess can identify the caller without
+// implementing its own auth.
+func (p *Plugin) StdioClients(uc *UserContext) []StdioServerClient {
 	var clients []StdioServerClient
 	for key, srv := range p.Manifest.MCPServers {
 		if srv.Type != "stdio" {
@@ -60,6 +70,16 @@ func (p *Plugin) StdioClients() []StdioServerClient {
 		env["DWS_PLUGIN_ROOT"] = p.Root
 		env["DWS_PLUGIN_DATA"] = filepath.Join(filepath.Dir(filepath.Dir(p.Root)), "data", p.Manifest.Name)
 
+		// Inject user identity so the subprocess knows who is calling.
+		if uc != nil {
+			if uc.UserID != "" {
+				env["DWS_USER_ID"] = uc.UserID
+			}
+			if uc.CorpID != "" {
+				env["DWS_CORP_ID"] = uc.CorpID
+			}
+		}
+
 		sc := transport.NewStdioClient(command, args, env)
 		clients = append(clients, StdioServerClient{Key: key, Client: sc})
 	}

internal/transport/client.go

@@ -212,7 +212,7 @@ func (r *ToolCallResult) UnmarshalJSON(data []byte) error {
 func defaultTransport() *http.Transport {
 	return &http.Transport{
 		DialContext: (&net.Dialer{
-			Timeout:   10 * time.Second,
+			Timeout:   3 * time.Second,
 			KeepAlive: 30 * time.Second,
 		}).DialContext,
 		TLSClientConfig:       &tls.Config{MinVersion: tls.VersionTLS12},