Responses API MCP Timeouts

API
1

Hi all,

I’m using the responses API along with the snowflake managed MCP server for a cortex agent. Sometimes the model will ask the agent for a complex query which could take time to process >60 seconds etc.

It seems that OpenAI times out long running mcp tool calls and I cannot see where this is configurable.

I understand the better approach is probably to design a tool to start a request and another to poll the agent but I don’t think snowflake currently supports this and if I can avoid it I would rather not build my own. Any help is much appreciated.

3 Likes
2

Even I am stuck here.

Please someone suggest something. Do let me know if you figure something out

3

Hey @Archer & or4, this isn’t ideal, agreed.

Did you get a chance to look at the developer doc Guide to Using the Responses API's MCP Tool? It includes some points on avoiding ballooning payloads and improving response times, which apply to long-running calls like this.

Let us know if that helps.

1 Like
4

I did have a look.

I don’t think the guide provides a way to handle long running tasks.

Can we somehow configure this timeout - either in the mcp server or in the responses api?

5

Yes I agree I don’t think this provides a solution, ideally we need a timeout config

6

Interesting. I know codex-cli has a tool_timeout_sec config option, which sets the maximum number of seconds the server will allow a tool invocation to run before timing out. If you don’t set it, it defaults to 60 seconds.

Let me see if I can find anything for long-running MCP tool calls on the Responses API.

1 Like
7

thanks for looking into this.

Did you find anything?