fix(mcp): retry Claude user config locks

src/utils/browser/mcp-installer.ts

@@ -4,10 +4,13 @@
 
 import * as fs from 'fs';
 import * as path from 'path';
-import * as lockfile from 'proper-lockfile';
 import { InstanceManager } from '../../management/instance-manager';
 import { getClaudeUserConfigPath } from '../claude-config-path';
 import { getCcsDir } from '../config-manager';
+import {
+  isClaudeUserConfigLockUnavailableError as isLockUnavailableError,
+  withClaudeUserConfigLock,
+} from '../claude-user-config-lock';
 
 const BROWSER_MCP_SERVER = 'ccs-browser-server.cjs';
 const BROWSER_MCP_SERVER_NAME = 'ccs-browser';
@@ -133,38 +136,6 @@ function writeClaudeUserConfig(configPath: string, config: ClaudeUserConfig): bo
   }
 }
 
-function withClaudeUserConfigLock<T>(configPath: string, callback: () => T): T {
-  const configDir = path.dirname(configPath);
-  const lockTarget = path.join(configDir, `${path.basename(configPath)}.ccs-lock`);
-  let release: (() => void) | undefined;
-
-  if (!fs.existsSync(configDir)) {
-    fs.mkdirSync(configDir, { recursive: true, mode: 0o700 });
-  }
-
-  if (!fs.existsSync(lockTarget)) {
-    fs.writeFileSync(lockTarget, '', { encoding: 'utf8', mode: 0o600 });
-  }
-
-  try {
-    release = lockfile.lockSync(lockTarget, { stale: 10000 }) as () => void;
-    return callback();
-  } finally {
-    if (release) {
-      try {
-        release();
-      } catch {
-        // Best-effort release.
-      }
-    }
-  }
-}
-
-function isLockUnavailableError(error: unknown): boolean {
-  const code = (error as NodeJS.ErrnoException | undefined)?.code;
-  return code === 'ELOCKED' || code === 'ENOTACQUIRED';
-}
-
 function removeManagedServerConfig(configPath: string): boolean {
   if (!fs.existsSync(configPath)) {
     return false;

src/utils/claude-user-config-lock.ts

@@ -0,0 +1,102 @@
+import * as fs from 'fs';
+import * as path from 'path';
+import * as lockfile from 'proper-lockfile';
+
+const DEFAULT_LOCK_STALE_MS = 10000;
+const DEFAULT_LOCK_RETRY_DELAY_MS = 200;
+const DEFAULT_LOCK_RETRY_TIMEOUT_MS = 10000;
+
+interface LockOptions {
+  staleMs?: number;
+  retryDelayMs?: number;
+  retryTimeoutMs?: number;
+}
+
+export function withClaudeUserConfigLock<T>(
+  configPath: string,
+  callback: () => T,
+  options: LockOptions = {}
+): T {
+  const configDir = path.dirname(configPath);
+  const lockTarget = path.join(configDir, `${path.basename(configPath)}.ccs-lock`);
+  const staleMs = options.staleMs ?? DEFAULT_LOCK_STALE_MS;
+  const retryDelayMs = options.retryDelayMs ?? DEFAULT_LOCK_RETRY_DELAY_MS;
+  const retryTimeoutMs = options.retryTimeoutMs ?? DEFAULT_LOCK_RETRY_TIMEOUT_MS;
+  let release: (() => void) | undefined;
+
+  if (!fs.existsSync(configDir)) {
+    fs.mkdirSync(configDir, { recursive: true, mode: 0o700 });
+  }
+
+  if (!fs.existsSync(lockTarget)) {
+    fs.writeFileSync(lockTarget, '', { encoding: 'utf8', mode: 0o600 });
+  }
+
+  try {
+    release = acquireClaudeUserConfigLock(
+      lockTarget,
+      configPath,
+      staleMs,
+      retryDelayMs,
+      retryTimeoutMs
+    );
+    return callback();
+  } finally {
+    if (release) {
+      try {
+        release();
+      } catch {}
+    }
+  }
+}
+
+export function isClaudeUserConfigLockUnavailableError(error: unknown): boolean {
+  const code = (error as NodeJS.ErrnoException | undefined)?.code;
+  return code === 'ELOCKED' || code === 'ENOTACQUIRED';
+}
+
+function acquireClaudeUserConfigLock(
+  lockTarget: string,
+  configPath: string,
+  staleMs: number,
+  retryDelayMs: number,
+  retryTimeoutMs: number
+): () => void {
+  const startedAt = Date.now();
+  let lastError: unknown;
+
+  for (;;) {
+    try {
+      return lockfile.lockSync(lockTarget, { stale: staleMs }) as () => void;
+    } catch (error) {
+      if (!isClaudeUserConfigLockUnavailableError(error)) {
+        throw error;
+      }
+      lastError = error;
+      if (Date.now() - startedAt >= retryTimeoutMs) {
+        throw buildLockTimeoutError(configPath, retryTimeoutMs, lastError);
+      }
+      sleepSync(retryDelayMs);
+    }
+  }
+}
+
+function sleepSync(ms: number): void {
+  if (ms <= 0) {
+    return;
+  }
+  Atomics.wait(new Int32Array(new SharedArrayBuffer(4)), 0, 0, ms);
+}
+
+function buildLockTimeoutError(
+  configPath: string,
+  timeoutMs: number,
+  cause: unknown
+): NodeJS.ErrnoException {
+  const causeCode = (cause as NodeJS.ErrnoException | undefined)?.code;
+  const error = new Error(
+    `Failed to acquire Claude user config lock for ${configPath} after ${timeoutMs}ms; another CCS process may be starting CLIProxy`
+  ) as NodeJS.ErrnoException;
+  error.code = causeCode === 'ENOTACQUIRED' ? 'ENOTACQUIRED' : 'ELOCKED';
+  return error;
+}

src/utils/image-analysis/mcp-installer.ts

@@ -4,7 +4,6 @@
 
 import * as fs from 'fs';
 import * as path from 'path';
-import * as lockfile from 'proper-lockfile';
 
 import { getCcsDir } from '../config-manager';
 import { getClaudeUserConfigPath } from '../claude-config-path';
@@ -12,6 +11,10 @@ import { info, warn } from '../ui';
 import { InstanceManager } from '../../management/instance-manager';
 import { installImageAnalysisPrompts } from './hook-installer';
 import { getImageAnalysisConfig } from '../../config/config-loader-facade';
+import {
+  isClaudeUserConfigLockUnavailableError as isLockUnavailableError,
+  withClaudeUserConfigLock,
+} from '../claude-user-config-lock';
 
 const IMAGE_ANALYSIS_MCP_SERVER = 'ccs-image-analysis-server.cjs';
 const IMAGE_ANALYSIS_MCP_RUNTIME = 'image-analysis-runtime.cjs';
@@ -121,38 +124,6 @@ function writeClaudeUserConfig(configPath: string, config: ClaudeUserConfig): bo
   }
 }
 
-function withClaudeUserConfigLock<T>(configPath: string, callback: () => T): T {
-  const configDir = path.dirname(configPath);
-  const lockTarget = path.join(configDir, `${path.basename(configPath)}.ccs-lock`);
-  let release: (() => void) | undefined;
-
-  if (!fs.existsSync(configDir)) {
-    fs.mkdirSync(configDir, { recursive: true, mode: 0o700 });
-  }
-
-  if (!fs.existsSync(lockTarget)) {
-    fs.writeFileSync(lockTarget, '', { encoding: 'utf8', mode: 0o600 });
-  }
-
-  try {
-    release = lockfile.lockSync(lockTarget, { stale: 10000 }) as () => void;
-    return callback();
-  } finally {
-    if (release) {
-      try {
-        release();
-      } catch {
-        // Best-effort release.
-      }
-    }
-  }
-}
-
-function isLockUnavailableError(error: unknown): boolean {
-  const code = (error as NodeJS.ErrnoException | undefined)?.code;
-  return code === 'ELOCKED' || code === 'ENOTACQUIRED';
-}
-
 export function hasImageAnalysisMcpServerInstalled(): boolean {
   return (
     fs.existsSync(getImageAnalysisMcpServerPath()) &&

src/utils/websearch/mcp-installer.ts

@@ -12,6 +12,10 @@ import { InstanceManager } from '../../management/instance-manager';
 import { installWebSearchHook } from './hook-installer';
 import { appendWebSearchTrace } from './trace';
 import { getWebSearchConfig } from '../../config/config-loader-facade';
+import {
+  isClaudeUserConfigLockUnavailableError as isLockUnavailableError,
+  withClaudeUserConfigLock,
+} from '../claude-user-config-lock';
 
 const WEBSEARCH_MCP_SERVER = 'ccs-websearch-server.cjs';
 const WEBSEARCH_MCP_SERVER_NAME = 'ccs-websearch';
@@ -118,47 +122,63 @@ function removeManagedServerConfig(configPath: string): boolean {
     return false;
   }
 
-  const config = readClaudeUserConfig(configPath);
-  if (config === null) {
-    if (process.env.CCS_DEBUG) {
-      console.error(warn(`Malformed Claude config prevents MCP cleanup: ${configPath}`));
-    }
-    return false;
-  }
-
-  const existingServers =
-    config.mcpServers && typeof config.mcpServers === 'object' && !Array.isArray(config.mcpServers)
-      ? { ...(config.mcpServers as Record<string, unknown>) }
-      : {};
-
-  if (!(WEBSEARCH_MCP_SERVER_NAME in existingServers)) {
-    return false;
-  }
-
-  delete existingServers[WEBSEARCH_MCP_SERVER_NAME];
-
-  const nextConfig: ClaudeUserConfig = { ...config };
-  if (Object.keys(existingServers).length === 0) {
-    delete nextConfig.mcpServers;
-  } else {
-    nextConfig.mcpServers = existingServers;
-  }
-
   try {
-    writeClaudeUserConfig(configPath, nextConfig);
-    if (process.env.CCS_DEBUG) {
-      console.error(info(`Removed WebSearch MCP config from ${configPath}`));
-    }
-    return true;
+    return withClaudeUserConfigLock(configPath, () => {
+      const config = readClaudeUserConfig(configPath);
+      if (config === null) {
+        if (process.env.CCS_DEBUG) {
+          console.error(warn(`Malformed Claude config prevents MCP cleanup: ${configPath}`));
+        }
+        return false;
+      }
+
+      const existingServers =
+        config.mcpServers &&
+        typeof config.mcpServers === 'object' &&
+        !Array.isArray(config.mcpServers)
+          ? { ...(config.mcpServers as Record<string, unknown>) }
+          : {};
+
+      if (!(WEBSEARCH_MCP_SERVER_NAME in existingServers)) {
+        return false;
+      }
+
+      delete existingServers[WEBSEARCH_MCP_SERVER_NAME];
+
+      const nextConfig: ClaudeUserConfig = { ...config };
+      if (Object.keys(existingServers).length === 0) {
+        delete nextConfig.mcpServers;
+      } else {
+        nextConfig.mcpServers = existingServers;
+      }
+
+      try {
+        writeClaudeUserConfig(configPath, nextConfig);
+        if (process.env.CCS_DEBUG) {
+          console.error(info(`Removed WebSearch MCP config from ${configPath}`));
+        }
+        return true;
+      } catch (error) {
+        if (process.env.CCS_DEBUG) {
+          console.error(
+            warn(
+              `Failed to remove WebSearch MCP config from ${configPath}: ${(error as Error).message}`
+            )
+          );
+        }
+        return false;
+      }
+    });
   } catch (error) {
-    if (process.env.CCS_DEBUG) {
-      console.error(
-        warn(
-          `Failed to remove WebSearch MCP config from ${configPath}: ${(error as Error).message}`
-        )
-      );
+    if (isLockUnavailableError(error)) {
+      appendWebSearchTrace('websearch_mcp_config_remove_skipped', {
+        reason: 'user_config_locked',
+        configPath,
+        error: (error as Error).message,
+      });
+      return false;
     }
-    return false;
+    throw error;
   }
 }
 
@@ -243,67 +263,86 @@ export function ensureWebSearchMcpConfig(): boolean {
   }
 
   const claudeUserConfigPath = getClaudeUserConfigPath();
-  const claudeUserConfigDir = path.dirname(claudeUserConfigPath);
-  const config = readClaudeUserConfig(claudeUserConfigPath);
-
-  if (config === null) {
-    appendWebSearchTrace('websearch_mcp_config_failed', { reason: 'malformed_user_config' });
-    if (process.env.CCS_DEBUG) {
-      console.error(warn('Malformed ~/.claude.json prevents WebSearch MCP provisioning'));
-    }
-    return false;
-  }
-
-  if (!fs.existsSync(claudeUserConfigDir)) {
-    fs.mkdirSync(claudeUserConfigDir, { recursive: true, mode: 0o700 });
-  }
-
-  const existingServers =
-    config.mcpServers && typeof config.mcpServers === 'object' && !Array.isArray(config.mcpServers)
-      ? (config.mcpServers as Record<string, unknown>)
-      : {};
-  const desiredServerConfig: ManagedWebSearchMcpConfig = {
-    type: 'stdio',
-    command: 'node',
-    args: [getWebSearchMcpServerPath()],
-    env: {},
-  };
-
-  const currentConfig = existingServers[WEBSEARCH_MCP_SERVER_NAME];
-  if (
-    typeof currentConfig === 'object' &&
-    currentConfig !== null &&
-    JSON.stringify(currentConfig) === JSON.stringify(desiredServerConfig)
-  ) {
-    appendWebSearchTrace('websearch_mcp_config_ready', { configPath: claudeUserConfigPath });
-    return true;
-  }
-
-  const nextConfig: ClaudeUserConfig = {
-    ...config,
-    mcpServers: {
-      ...existingServers,
-      [WEBSEARCH_MCP_SERVER_NAME]: desiredServerConfig,
-    },
-  };
 
   try {
-    writeClaudeUserConfig(claudeUserConfigPath, nextConfig);
-    appendWebSearchTrace('websearch_mcp_config_ready', { configPath: claudeUserConfigPath });
-    if (process.env.CCS_DEBUG) {
-      console.error(info(`Ensured WebSearch MCP config in ${claudeUserConfigPath}`));
-    }
-    return true;
-  } catch (error) {
-    appendWebSearchTrace('websearch_mcp_config_failed', {
-      reason: 'write_failed',
-      configPath: claudeUserConfigPath,
-      error: (error as Error).message,
+    return withClaudeUserConfigLock(claudeUserConfigPath, () => {
+      const config = readClaudeUserConfig(claudeUserConfigPath);
+
+      if (config === null) {
+        appendWebSearchTrace('websearch_mcp_config_failed', { reason: 'malformed_user_config' });
+        if (process.env.CCS_DEBUG) {
+          console.error(warn('Malformed ~/.claude.json prevents WebSearch MCP provisioning'));
+        }
+        return false;
+      }
+
+      const existingServers =
+        config.mcpServers &&
+        typeof config.mcpServers === 'object' &&
+        !Array.isArray(config.mcpServers)
+          ? (config.mcpServers as Record<string, unknown>)
+          : {};
+      const desiredServerConfig: ManagedWebSearchMcpConfig = {
+        type: 'stdio',
+        command: 'node',
+        args: [getWebSearchMcpServerPath()],
+        env: {},
+      };
+
+      const currentConfig = existingServers[WEBSEARCH_MCP_SERVER_NAME];
+      if (
+        typeof currentConfig === 'object' &&
+        currentConfig !== null &&
+        JSON.stringify(currentConfig) === JSON.stringify(desiredServerConfig)
+      ) {
+        appendWebSearchTrace('websearch_mcp_config_ready', { configPath: claudeUserConfigPath });
+        return true;
+      }
+
+      const nextConfig: ClaudeUserConfig = {
+        ...config,
+        mcpServers: {
+          ...existingServers,
+          [WEBSEARCH_MCP_SERVER_NAME]: desiredServerConfig,
+        },
+      };
+
+      try {
+        writeClaudeUserConfig(claudeUserConfigPath, nextConfig);
+        appendWebSearchTrace('websearch_mcp_config_ready', { configPath: claudeUserConfigPath });
+        if (process.env.CCS_DEBUG) {
+          console.error(info(`Ensured WebSearch MCP config in ${claudeUserConfigPath}`));
+        }
+        return true;
+      } catch (error) {
+        appendWebSearchTrace('websearch_mcp_config_failed', {
+          reason: 'write_failed',
+          configPath: claudeUserConfigPath,
+          error: (error as Error).message,
+        });
+        if (process.env.CCS_DEBUG) {
+          console.error(warn(`Failed to update ~/.claude.json: ${(error as Error).message}`));
+        }
+        return false;
+      }
     });
-    if (process.env.CCS_DEBUG) {
-      console.error(warn(`Failed to update ~/.claude.json: ${(error as Error).message}`));
+  } catch (error) {
+    if (isLockUnavailableError(error)) {
+      appendWebSearchTrace('websearch_mcp_config_failed', {
+        reason: 'user_config_locked',
+        configPath: claudeUserConfigPath,
+        error: (error as Error).message,
+      });
+      if (process.env.CCS_DEBUG) {
+        console.error(
+          warn(
+            `WebSearch MCP config skipped because ${claudeUserConfigPath} is locked by another process`
+          )
+        );
+      }
+      return false;
     }
-    return false;
+    throw error;
   }
 }
 

tests/unit/utils/image-analysis/mcp-installer.test.ts

@@ -194,26 +194,54 @@ describe('ensureImageAnalysisMcp', () => {
     expect(lockSpy.mock.calls[0]?.[0]).toBe(path.join(tempHome as string, '.claude.json.ccs-lock'));
   });
 
-  it('returns false instead of throwing when ~/.claude.json is already locked', () => {
+  it('retries ~/.claude.json lock acquisition before provisioning', () => {
+    setupTempHome();
+    writeEnabledConfig();
+    const claudeUserConfigPath = path.join(tempHome as string, '.claude.json');
+    fs.writeFileSync(claudeUserConfigPath, '{}\n', 'utf8');
+
+    let calls = 0;
+    spyOn(lockfile, 'lockSync').mockImplementation(() => {
+      calls++;
+      if (calls === 1) {
+        const error = new Error('Lock file is already being held') as NodeJS.ErrnoException;
+        error.code = 'ELOCKED';
+        throw error;
+      }
+      return (() => undefined) as ReturnType<typeof lockfile.lockSync>;
+    });
+
+    expect(ensureImageAnalysisMcp()).toBe(true);
+    expect(calls).toBe(2);
+
+    const config = JSON.parse(fs.readFileSync(claudeUserConfigPath, 'utf8')) as {
+      mcpServers: Record<string, unknown>;
+    };
+    expect(config.mcpServers[getImageAnalysisMcpServerName()]).toEqual(getManagedConfig());
+  });
+
+  it('returns false instead of throwing when ~/.claude.json stays locked', () => {
     setupTempHome();
     writeEnabledConfig();
 
     const claudeUserConfigPath = path.join(tempHome as string, '.claude.json');
     fs.writeFileSync(claudeUserConfigPath, '{}\n', 'utf8');
-    fs.writeFileSync(path.join(tempHome as string, '.claude.json.ccs-lock'), '', 'utf8');
 
-    const release = lockfile.lockSync(path.join(tempHome as string, '.claude.json.ccs-lock'), {
-      stale: 10000,
-    }) as () => void;
+    let now = 0;
+    spyOn(Date, 'now').mockImplementation(() => {
+      now += 10001;
+      return now;
+    });
+    spyOn(lockfile, 'lockSync').mockImplementation(() => {
+      const error = new Error('Lock file is already being held') as NodeJS.ErrnoException;
+      error.code = 'ELOCKED';
+      throw error;
+    });
 
-    try {
-      let result: boolean | undefined;
-      expect(() => {
-        result = ensureImageAnalysisMcp();
-      }).not.toThrow();
-      expect(result).toBe(false);
-    } finally {
-      release();
-    }
+    let result: boolean | undefined;
+    expect(() => {
+      result = ensureImageAnalysisMcp();
+    }).not.toThrow();
+    expect(result).toBe(false);
   });
 });

tests/unit/utils/websearch/mcp-installer.test.ts

@@ -2,6 +2,7 @@ import { afterEach, describe, expect, it, mock, spyOn } from 'bun:test';
 import * as fs from 'fs';
 import * as os from 'os';
 import * as path from 'path';
+import * as lockfile from 'proper-lockfile';
 import { getHookPath } from '../../../../src/utils/websearch/hook-config';
 import {
   ensureWebSearchMcp,
@@ -33,7 +34,15 @@ describe('ensureWebSearchMcp', () => {
     fs.mkdirSync(ccsDir, { recursive: true });
     fs.writeFileSync(
       path.join(ccsDir, 'config.yaml'),
-      ['version: 12', 'websearch:', '  enabled: true', '  providers:', '    duckduckgo:', '      enabled: true', ''].join('\n'),
+      [
+        'version: 12',
+        'websearch:',
+        '  enabled: true',
+        '  providers:',
+        '    duckduckgo:',
+        '      enabled: true',
+        '',
+      ].join('\n'),
       'utf8'
     );
   }
@@ -95,6 +104,19 @@ describe('ensureWebSearchMcp', () => {
     expect(config.mcpServers[getWebSearchMcpServerName()]).toEqual(getManagedConfig());
   });
 
+  it('serializes ~/.claude.json updates with a file lock', () => {
+    setupTempHome();
+    writeEnabledConfig();
+    const claudeUserConfigPath = path.join(tempHome as string, '.claude.json');
+    fs.writeFileSync(claudeUserConfigPath, '{}\n', 'utf8');
+
+    const lockSpy = spyOn(lockfile, 'lockSync');
+
+    expect(ensureWebSearchMcp()).toBe(true);
+    expect(lockSpy).toHaveBeenCalled();
+    expect(lockSpy.mock.calls[0]?.[0]).toBe(path.join(tempHome as string, '.claude.json.ccs-lock'));
+  });
+
   it('preserves the existing ~/.claude.json permissions when provisioning WebSearch MCP', () => {
     setupTempHome();
     writeEnabledConfig();
@@ -170,7 +192,11 @@ describe('ensureWebSearchMcp', () => {
       'utf8'
     );
 
+    const lockSpy = spyOn(lockfile, 'lockSync');
+
     expect(uninstallWebSearchMcp()).toBe(true);
+    expect(lockSpy).toHaveBeenCalled();
+    expect(lockSpy.mock.calls[0]?.[0]).toBe(path.join(tempHome as string, '.claude.json.ccs-lock'));
     expect(fs.existsSync(getWebSearchMcpServerPath())).toBe(false);
 
     const globalConfig = JSON.parse(fs.readFileSync(claudeUserConfigPath, 'utf8')) as {