Multiple tool calls in a single Gemini Response breaks the agent.

[exi]

What happened?

I have a bunch of tools defined for my gemini-2.0-flash-exp agent.

When it uses one tool at a time, everything is fine. However once it starts using more than one tool in a single reply, the very next message will fail immediately with an API error:

Googling suggests that the Google API expected the exact number of tool results back in the next request as it requested. Not doing this will lead to this error I think.

Version Information

ddb2141

Steps to Reproduce

1. Prompt the agent with something that would require multiple tool calls.
2. Have the agent respond with a list of tool calls.
3. Error appears on the next request.

What browsers are you seeing the problem on?

Firefox

Relevant log output

LibreChat         | 2025-05-03 21:32:38 error: [api/server/controllers/agents/client.js #sendCompletion] Unhandled error type [GoogleGenerativeAI Error]: Error fetching from https://generativelanguage.googleapis.com/v1beta/models/gemini-2.0-flash-exp:streamGenerateContent?alt=sse: [400 Bad Request] Please ensure that the number of function response parts is equal to the number of function call parts of the function call turn.

Screenshots

Code of Conduct

• I agree to follow this project's Code of Conduct

[danny-avila]

We are including the same amount of functions and it would work with any other provider. It seems like a common issue across many apps with Google since they have some strictness with the way function responses are handled. It can’t be reliably reproduced on my end though.

Some more insight on this:
firebase/firebase-ios-sdk#14022

[danny-avila]

I realized this was reported before and it ended up being the reverse proxy at issue (one-api). If you are using a reverse proxy, I would report it to the relevant project.

I still can't reproduce this issue.

After doing more research, it seems this error can also be returned by the Google API when the payload starts with an AI message, and I'll investigate that on my end, if I can cause it that way.

[exi]

I have a repro!

The error seems to occur once i start using two different MCP servers in one message.

Example of one MCP server after another:

Example when doing it in one step instead of two:

[exi]

@danny-avila There you go: https://gist.github.com/exi/5a9fe130a1dba2870c96c7b46c1cf83a

[danny-avila]

thanks! taking a look soon.

[exi]

@danny-avila upstream has fixed the issue. Can this be closed? Is the fix in master yet?

[danny-avila]

Hi @exi thanks for following up. I will include the fix today and then close this

[danny-avila]

Merging this in main soon: #7525