MultiServerMCPClient does not work with Mistral AI

[nicolas-dutertry]

I have tried to run the sample from the doc with a simple mcp calculator (https://docs.langchain.com/oss/python/langchain/mcp).
It works with claude or gpt but it does not work with mistral.
I have made other tests with MultiServerMCPClient and Mistral but I always have the same error:

httpx.HTTPStatusError: Error response 422 while fetching https://api.mistral.ai/v1/chat/completions: {"detail":[{"type":"string_type","loc":["body","messages",2,"tool","content","str"],"msg":"Input should be a valid string","input":[{"type":"text","text":"8","id":"lc_dece2988-cfd4-4169-8fba-2c72c3d114af"}]},{"type":"extra_forbidden","loc":["body","messages",2,"tool","content","list[tagged-union[TextChunk,ImageURLChunk,DocumentURLChunk,ReferenceChunk,BBoxChunk,FileURLChunk,AudioChunk,FileChunk,ThinkChunk]]",0,"text","id"],"msg":"Extra inputs are not permitted","input":"lc_dece2988-cfd4-4169-8fba-2c72c3d114af"},{"type":"string_type","loc":["body","messages",3,"tool","content","str"],"msg":"Input should be a valid string","input":[{"type":"text","text":"96","id":"lc_02676c04-2c51-4365-96e5-50394ec47bdc"}]},{"type":"extra_forbidden","loc":["body","messages",3,"tool","content","list[tagged-union[TextChunk,ImageURLChunk,DocumentURLChunk,ReferenceChunk,BBoxChunk,FileURLChunk,AudioChunk,FileChunk,ThinkChunk]]",0,"text","id"],"msg":"Extra inputs are not permitted","input":"lc_02676c04-2c51-4365-96e5-50394ec47bdc"}]}

Here is the code sample :

client = MultiServerMCPClient(
    {
        "math": {
            "transport": "stdio",  # Local subprocess communication
            "command": "python",
            # Absolute path to your math_server.py file
            "args": ["/test-mcp/src/test_mcp/math_server.py"],
        }
    }
)
tools = await client.get_tools()
agent = create_agent(
    "mistralai:mistral-large-latest",
    tools
)
math_response = await agent.ainvoke(
    {"messages": [{"role": "user", "content": "what's (3 + 5) x 12?"}]}
)

[darkydash]

Looks like tools under MultiServerMCPClient always runs with response_format = "content_and_artifact" option (hardcoded here https://github.com/langchain-ai/langchain-mcp-adapters/blob/main/langchain_mcp_adapters/tools.py#L431). Mistral does not support that. But if you change it explicitly it makes it work
E.g.: tools[0].response_format = "content"

[darkydash]

Also seems weird to me that content returns tuple but content_and_artifact - doesn't. Bug?

[nicolas-dutertry]

Adding tools[0].response_format = "content" doesn't fix the issue.

The issue seems to be related to attribute "id" added in the content of ToolMessage. For instance:
"content" : [{"type":"text","text":"96","id":"lc_e6adc0f5-074f-4974-9894-f601e86884ef"}]

I have monkey patched _convert_mcp_content_to_lc_block to remove "id", and it works:

original_convert_mcp_content_to_lc_block = langchain_mcp_adapters.tools._convert_mcp_content_to_lc_block

def patched_convert_mcp_content_to_lc_block(
    content) :
    result = original_convert_mcp_content_to_lc_block(content)
    if('id' in result):
        del result['id']
    return result

langchain_mcp_adapters.tools._convert_mcp_content_to_lc_block = patched_convert_mcp_content_to_lc_block

[darkydash]

For me it works as follows