Cursor MCP server subsystem is totally SNAFU - Please can you prioritize some fixes!

[Chris_Drake]

Where does the bug appear (feature/product)?

Cursor IDE

Describe the Bug

• Me and others have reported, days ago, a critical issue with tool calls going to the wrong MCP server, which remains unfixed.
• The IDE still makes multiple unnecessary simultaneous connections to non-stdio MCP servers
• If any server goes down, cursor drops all connections to all servers
• Whether or not cursor works after changing the mcp.json file is 50%/50% half -the time it says “Error show output” (when there was nothing wrong with the server at all)
• When it errors, it stops re-trying to connect again - so any error, no matter how short, cuts off the server indefinitely
• The cursor-agent wsl tool cannot read or use the cursor agent mcp.json file (doesn’t authenticate like the IDE does)
• and that’s just off-the-top-of-my head things I can remember right now!

Steps to Reproduce

All these bugs have been reported many times in other bug reports

Expected Behavior

Someone working on the IDE needs to fire up “Claude 4.5”, tell it what I said above, and ask it to fix the code.
None of those are hard problems, and I’ve given plenty of help in my other bug reports to find and fix them.

Operating System

Windows 10/11
Linux

Current Cursor Version (Menu → About Cursor → Copy)

All to current, IDE and CLI as well.

Additional Information

One of these is a critical failure - sending valid tool calls to a wrong MCP server can be catastrophic and even illegal if PII is involved.

Does this stop you from using Cursor

Yes - Cursor is unusable

[deanrie]

Hey, thanks for the report. I can confirm that several of these issues are already tracked and being worked on by the engineering team.

Confirmed and tracked issues:

1. Tool calls going to the wrong server
   • Temporary workaround: run one MCP server at a time, or rename tools to avoid conflicts - CRITICAL BUG: Cursor routing tool calls to wrong MCP servers
2. Multiple simultaneous connections to non-stdio servers
   • Race condition causing duplicate containers to be created - MCP Docker containers fail with name conflicts after Oct 29 update
3. Auto-reconnection errors
   • Multiple parallel reconnection attempts trigger “Error - Show Output” - Auto-reconnection to MCP servers is broken (runs multiple times in parallel, then fails). on/off works OK
4. MCP authentication for cursor-agent CLI
   • Known limitation in headless/CI environments, there are workarounds but they are manual - MCP servers are not recognized with cursor-cli in a CI environment

We need more information on:

• “Drops all connections if one server fails”
  • Could you share logs that show this behavior?
• “50/50 Error show output” when changing mcp.json
  • Is this the same as issue #3 above, or a different one?
  • If it’s different, please share the MCP logs from ~/Library/Application Support/Cursor/logs/
• “Stops re-trying to connect indefinitely”
  • Request IDs or logs would help us understand what’s happening more clearly

Also, please share your exact Cursor version:

• Help > About Cursor > Copy.

The team understands these are high-priority issues, especially the tool routing bug, given the security concerns you mentioned.

[Chris_Drake]

Your comment here seems off-topic?

Multiple simultaneous connections to non-stdio servers
Race condition causing duplicate containers to be created - MCP Docker containers fail with name conflicts after Oct 29 update

docker is a local thing, non-stdio has nothing to do with docker.

As best I can tell, there’s some bug in your loop - if there’s 4 MCP servers, it makes 4 new connections to each one of those 4 servers - or something like that.

[Chris_Drake]

@deanrie

“Drops all connections if one server fails”

• Could you share logs that show this behavior?

• “Drops all connections if one server fails”
  • Could you share logs that show this behavior?

Logs are attached.

(secondary bug report: no way to copy out of that box Ctrl-C and right-click copy does nothing.)

Version: 2.0.77 (system setup)
VSCode Version: 1.99.3
Commit: ba90f2f88e4911312761abab9492c42442117cf0
Date: 2025-11-13T23:10:43.113Z
Electron: 37.7.0
Chromium: 138.0.7204.251
Node.js: 22.20.0
V8: 13.8.258.32-electron.0
OS: Windows_NT x64 10.0.26100

I was connected to 2 MCP servers.

The logs show:

a) after I stopped one of the servers, cursor dropped all connections to all server (i.e. the Administrator server got dropped and reconnected for no reason)
The highlighted portion of the log lines show what came up after I stopped the server.

b) I restarted the server I stopped - notice how you can still see that your interface says an error message, even though the server is back up again.

Update: (another bug report)
The interface “lies” - it says “Error…” etc and a red dot, even though the server and tools are connected and working fine.