500 error from gemini-2.5-pro model with 10 mcp servers consisting of 52 tools

[sparshpml1212]

I'm seeing 500 error from gemini when I try connecting with 10 mcp servers consisting of 52 tools.
There's no issue with one of the mcp servers, as I have tried removing and attaching tools one by one.

The possible reasons could be context length or something but gemini 2.5 pro provides 1M input token length so it should not be an issue.

Also this setup works perfectly fine with gemini-2.5-flash.

I'm using MultiServerMCPClient from lanchain-mcp-adaptors.

[msaadg]

Seems like an issue with the Gemini side. Maybe there's a lower limit with the pro models (not sure though). Can you share a reproducible example for the error so I can investigate further?

[sparshpml1212]

Seems like an issue with the Gemini side. Maybe there's a lower limit with the pro models (not sure though). Can you share a reproducible example for the error so I can investigate further?

Steps To reproduce:

1. Create a Simple graph in this format -> start -> model call -> tool calls (conditional edge) -> final model call -> END
2. Make MCP connection using MultiServerMCPClient from langchain-mcp-adaptors.
3. Take around 10 mcp servers(you may just use mocks) and have 5-5 tools in each mcp server and make sure to have large descriptions(docstrings) for these tools
4. Try the setup with gemini-2.5-pro and gemini-2.5-flash

Observation: this setup works well with flash model but not with pro model.