Set longer request times for Inspector client

[luiztauffer]

Describe the bug
I get Error: Request timed out whenever I send requests to tools that take a little bit longer to process.

To Reproduce
Just use any MCP tool that takes a longer time to finish, e.g.

from time import sleep
from mcp.server.fastmcp import FastMCP

mcp = FastMCP("myserver", version="0.1.0", request_timeout=300)

@mcp.tool()
def return_string(query: str) -> str:
    sleep(10)
    return f"Query: {query}"

then run it with:

mcp dev main.py

Expected behavior
I would expect the client to wait for the response up to a limit, but I don't know how to set that limit in the Inspector.

[kwnath]

Not sure if this is a bug but the default request timeout is 10s (https://github.com/modelcontextprotocol/inspector/blob/master/client/src/lib/hooks/useConnection.ts#L23) which is why you're running into the error you've attached.

You've got a few options here:

1. If you're running this locally I'd suggest just increasing the timeout to whatever you feel makes sense.
2. The alternative is open a PR to increase the timeout but I think 10s is pretty sensible.

The former probably makes more sense for you 😄

[luiztauffer]

hi @kwnath yeah I agree this is not a bug, the tag came automatically I don't know why
thank you for the suggestions! Yes I know I could just change that constant and run it from source, but I think it would be useful to have it as a configuration option directly in the UI, since many MCPs that do a bit heavier processing will certainly require a longer time.
If you think this is an useful feature I can leave this Issue open, if there's no interest in doing that modification I can close it

[jspahrsummers]

Seems like a valid enhancement request, thanks! Will leave it open.

[Lakphy]

I have the same problem, in my tool I may also make some llm calls, making the whole process take longer, which will make it difficult to debug if Inspector client limits the timeout to 10 seconds.

[liuzhuang-bud]

I have the same problem too, in my tool I may also make some llm calls,

[hp1998318]

我更改了超长时间后 本地调试 一直显示端口被占用了！

node:events:495 throw er; // Unhandled 'error' event ^ Error: listen EADDRINUSE: address already in use :::3000 �[90m at Server.setupListenHandle [as _listen2] (node:net:1817:16)�[39m �[90m at listenInCluster (node:net:1865:12)�[39m �[90m at Server.listen (node:net:1953:7)�[39m at Function.listen (/Users/bytedance/code/inspector/node_modules/�[4mexpress�[24m/lib/application.js:635:24) at �[90mfile:///Users/bytedance/code/inspector/server/�[39mbuild/index.js:149:24 Emitted 'error' event on Server instance at: �[90m at emitErrorNT (node:net:1844:8)�[39m �[90m at process.processTicksAndRejections (node:internal/process/task_queues:82:21)�[39m { code: �[32m'EADDRINUSE'�[39m, errno: �[33m-48�[39m, syscall: �[32m'listen'�[39m, address: �[32m'::'�[39m, port: �[33m3000�[39m } Node.js v18.20.3

[Lurrobert]

10 seconds is definitely a bug. Please fix. it is literally 1 prompt away

[olaservo]

This PR from @TornjV restored a way to override the timeout using a query string parameter: #164

I'm going to keep this issue open since I think we should still have a UI option for usability's sake, but hopefully this option will be easier than having to change the app code to override the default.