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.

[Cloud-Iris]

Hello? Is the problem still unresolved? Maybe someone could open a PR to change the frontend by giving a textarea for user to change.

[olaservo]

Hi @Cloud-Iris , for now there is just the option to override using the URL param (you can take a look at the PR linked above for more details).

Adding it to the UI is still open as an enhancement. It hasn't been on the top of my list personally since there's a lot of other issues to address in this and other MCP repos, but feel free to open a PR and it will typically get reviewed within a few days.