LangChain integration doesn't work with MCP tools and Anthropic models

[snersisyan-lilly]

Hello,

When using LangChain with ChatDatabricks that uses any Anthropic model hosted on Databricks, it throws an error. This is quite critical, as the default Databricks App template called "agent-langgraph" doesn't work with Anthropic models out of box.

Here is minimal reproducible example:

from langchain.agents import create_agent
from langchain_mcp_adapters.client import MultiServerMCPClient
from databricks_langchain import ChatDatabricks

async def main_langchain():
    client = MultiServerMCPClient(
        {
            "my_server": {
                "transport": "http",
                "url": "https://pubmed.mcp.claude.com/mcp",
            }
        }
    )

    tools = await client.get_tools()

    agent = create_agent(
        model=ChatDatabricks(endpoint="databricks-claude-sonnet-4-6"),
        tools=tools,
    )

    result = await agent.ainvoke(
        {"messages": [{"role": "user", "content": "Find papers by Lalo Salamanca (PMIDs)."}]},
        config={"configurable": {"thread_id": "1"}},
    )
    return result


await main_langchain()

Throws error:

BadRequestError: Error code: 400 - {'error_code': 'BAD_REQUEST', 'message': '{"message":"messages.2.content.0.tool_result.content.0.text.id: Extra inputs are not permitted"}'}
Error(message=Cannot find child , errno=no such file or directory, statusCode=0)
[Trace ID: 00-9ff9b3d26db248cf9efe48e624e14137-8d6db60c4bd9c9ea-00]
File <command-7075019236347811>, line 27
     20     result = await agent.ainvoke(
     21         {"messages": [{"role": "user", "content": "Find papers by Lalo Salamanca (PMIDs)."}]},
     22         config={"configurable": {"thread_id": "1"}},
     23     )
     24     return result
---> 27 await main_langchain()

Same if I use DatabricksMCPServer, DatabricksMultiServerMCPClient.

Package versions:
langchain = 1.3.10
databricks-langchain = 0.20.0

[snersisyan-lilly]

Just in case, everything works e.g. with GPT models.