How to Pass API Token to @mcp.tool in langchain-mcp-adapters When Using LangGraph?

[wen020]

Body:
Hi,

I'm using LangGraph to call langchain-mcp-adapters and set up an MCP server. I have a question about securely passing authentication tokens (such as an API key) into an MCP tool function.

For example, given this MCP tool:

@mcp.tool()
async def get_weather(location: str) -> str:
    """Get weather for location."""
    return "It's always sunny in New York"

How can I pass an API token (used to call an external weather API) into this function? What's the recommended way to make sure the token is accessible inside get_weather? Should I use function parameters, environment variables, context, or something else?

If possible, could you provide a concrete code example of how to do this with langchain-mcp-adapters and LangGraph?

Thank you!

[vbarda]

If the API token is static, and not changing, you can simply load it inside the tool function or pass it via MCP context (requires defining app lifespan in your server) https://github.com/modelcontextprotocol/python-sdk?tab=readme-ov-file#server

If you want to pass it dynamically, at runtime, it is not supported by MCP at the moment (see issue here modelcontextprotocol/python-sdk#375). Once that's supported, we will add it in the library as well and make it compatible with standard LangChain/LangGraph practices.

In LangGraph this is usually done via an injected config argument: https://langchain-ai.github.io/langgraph/agents/context/#tools

[vbarda]

For static keys, it would simply be this:

@mcp.tool()
async def get_weather(location: str) -> str:
    """Get weather for location."""
    my_api_key = os.environ["WEATHER_API_KEY"]
    weather_response = get_weather(location, my_api_key)
    return weather_response