Claude Desktop: Built-in MCP servers 'Claude in Chrome' and 'mcp-registry' fail to connect (Failed to get account UUID)
Description
Claude Desktop v1.1.3647 on macOS consistently fails to connect its two built-in MCP servers: "Claude in Chrome" and "mcp-registry". These errors appear as toast notifications every time extensions are installed or conversations start. User-installed extensions (Control Chrome, iMessage, Desktop Commander) all connect successfully.
Error Details
Claude in Chrome
[chrome-mcp] Failed to get account UUID
[Claude in Chrome] No user ID available after Xms
chrome_bridge_connection_failed | error_type: 'no_user_id'
[Claude in Chrome] Bridge reconnecting in 3000ms (attempt 2)
The bridge registers successfully on startup:
[Chrome MCP] Bridge feature flag: chrome_ext_bridge_enabled_desktop=true, IS_NEST_BUILD=false
[Claude in Chrome] MCP server registered
But when a conversation requests a connection, the server never launches. The main.log shows MCP Server connection requested for: Claude in Chrome but no corresponding Launching MCP Server entry appears (unlike user-installed extensions which do launch).
mcp-registry
Same pattern: MCP Server connection requested for: mcp-registry appears in logs but the server never launches. No explicit error logged.
Environment
- Claude Desktop: v1.1.3647 (commit 8f7c53e009d62927d53b39524d8fd204b2bb8b10)
- macOS: Darwin 25.3.0 (Sequoia)
- Chrome extension: Installed (ID: fcoeoabgfenejglbffodgkkbkcdhcgfn, v1.0.54)
- Chrome native host: Working (socket created at
/tmp/claude-mcp-browser-bridge-*/, listening for connections) - Node.js: v25.4.0
Key Observations
- OAuth works for conversations — the user is signed in and can chat normally.
[oauth] using cached token for orgId=...appears in logs. - The account UUID retrieval specifically fails — even though OAuth tokens are valid, the Chrome MCP bridge component cannot get the account UUID.
- Native messaging infrastructure is healthy — the
chrome-native-hostbinary creates sockets, listens, but Claude Desktop never connects to them (no "Accepted new MCP connection" in native host logs). - User-installed extensions work fine — Control Chrome, iMessage, and Desktop Commander all show "Server started and connected successfully" in their logs.
- The error is persistent — occurs across multiple app restarts, sign out/sign in cycles, and Chrome extension re-pairing.
- Wait times vary wildly — from 43ms to 992272ms (16+ minutes), suggesting the bridge genuinely waits for auth but never gets it.
Steps to Reproduce
- Install Claude Desktop v1.1.3647 on macOS
- Sign in to Claude Desktop
- Install the Claude Chrome extension and pair it
- Install any extension from the marketplace (e.g., Control Chrome)
- Observe toast notifications: "Could not connect to MCP server mcp-registry" and "Could not connect to MCP server Claude in Chrome"
What Was Tried
- Signing out and back into Claude Desktop
- Re-pairing the Chrome extension
- Restarting both Chrome and Claude Desktop
- Reinstalling extensions
None of these resolved the issue.
Expected Behavior
Built-in MCP servers should connect without errors, or at minimum fail silently without showing error toasts to the user (since user-installed extensions work fine).
10 Comments
Found 3 possible duplicate issues:
This issue will be automatically closed as a duplicate in 3 days.
🤖 Generated with Claude Code
Experiencing this on February 21 and 22, 2026, right after an update to Claude Desktop app. Just updated evening of 2/22/26 and it's still broken. Same error messages: can't connect to mcp-registry and Claude in Chrome. Workflow requires local MCP access.
Yup. Experiencing same issues here as well. I am using Claude Desktop for MCP servers in Docker Desktop. And everytime its this error Could not connect to MCP server mcp-registry. And I have tried all possible solutions,
Same issue happend in desktop and chrome . Could not connect to mcp server mcp -register
Same for me, but windows
Same here.
same here.
same here
Same error, uninstall and reinstall didn't fit. Restart didn't fix. Checking other troubleshooting methods soon.
Did upgrade to 1.1.4088 and potentially the error went away, but I am getting the "response is taking longer" constant retries.
After the fix announcement, the app updated to 1.1.4088 and the problem seems indeed fixed. No more glitches.