Remote MCP connections fails every few minutes

[yadgire7]

Where does the bug appear (feature/product)?

Cursor IDE

Describe the Bug

I see this message: ‘Error - show output’ on the Tools and MCP tab after every few minutes of connecting to a remote MCP. The logs show - ‘Client error for command fetch failed’.
I am on Version: 2.2.23

Steps to Reproduce

connect to a remote MCP - use it - let it be idle for some time - go to the Tools and MCP tab

Operating System

MacOS

Current Cursor Version (Menu → About Cursor → Copy)

Version: 2.2.23
VSCode Version: 1.105.1

Does this stop you from using Cursor

Sometimes - I can sometimes use Cursor

[yadgire7]

found a relevant issue reported earlier: 1st sse mcp connection ALWAYS fails on windows (reconnect always works)

[Colin]

Hey there.

Thanks for the report. I can reproduce this pretty well with an internal MCP server that we use.

My error seems to always be

2025-12-15 18:10:28.825 [error] Client error for command SSE error: TypeError: terminated: Body Timeout Error

What error do you get? Would be great if you could share more from the logs.

[roqqit]

Relatedly why does Cursor need to hit the MCP server every few minutes? Do MCP tools really change that often?

It seems more reasonable that refetching tools should only occur when the user is interfacing with Cursor. If I have it minimized, it’s not like an update to the tools will matter to me until I start typing into the chat box again.

[yadgire7]

Thanks for the quick response.
This is what I see:

2025-12-15 12:16:36.910 [error] Client error for command fetch failed

2025-12-15 12:16:36.910 [error] Client error for command fetch failed

[Colin]

Hey @yadgire7

I found an existing thread:

Refresh_tokens are not used

Where does the bug appear (feature/product)? Cursor IDE Describe the Bug It looks like refresh_token isn’t used when using oauth We have the following setup of the MCP for test: MCP server built on top of Apollo MCP locally deployed oauth server (proxying requests to auth0) auth0 configured (for test) to give 1-minute valid access tokens refresh_token is supplied with access_token After login the connection works for 1 minute, as expected, then at some point we hit “unauthorized” and token …

And linked this thread to the corresponding internal ticket.

Thanks again for the report.

[yadgire7]

Thanks @Colin
The reply says:

This looks like a bug in the OAuth refresh flow for MCP servers.

Is it on Cursor side?

cc: @deanrie

[Colin]

Yes, exactly.

[yadgire7]

Okay. Can we expect the fix in coming stable versions?