Passing Authentication through LangFlow into Tools API calls

[andrewkcarter]

We have API's exposed as tools wrapped by an MCP server. Does LangFlow provide a mechanism to forward headers into the tool calls? The use case is to trigger a flow via API, and include the necessary headers in the flow invocation. These headers need to be included in the tool calls in order to use the tools.

For reference, the reference Python MCP server recently adopted changes to permit this, provided the caller is sending the data. modelcontextprotocol/python-sdk#816

[alexfhub]

@andrewkcarter Same issue here. Did you find a solution?
I tried using header forwarding in the MCP Connection, but no luck.

[andrewkcarter]

No, unfortunately no solution at this moment. We may need to offer a PR introducing this.

[alexfhub]

I just got it working!
Here’s how I passed the header into the MCP server code:

from mcp.server.fastmcp import FastMCP
import requests

mcp = FastMCP("myserver")

#Passing the authorization token through
def get_auth_header() -> dict:
    try:
        ctx = mcp.get_context()
        request = getattr(ctx.request_context, "request", None)
        if request:
            auth = request.headers.get("authorization")
            if auth:
                return {"Authorization": auth}
    except Exception as e:
        print(f"Auth error: {e}")
    return {}

#A tool that includes an API method with header-based authorization
@mcp.tool()
def some_api_method() -> str:
    """
    API description
    """
    url = "https://api.example.com/do"
    headers = get_auth_header()
    ...

After that, it correctly picked up the header I specified in the Langflow UI under MCP Connection settings:

And to pass that header via the Langflow API itself, this version worked:

# Some external app for interacting with the agent inside a LangFlow workflow

...

import requests
import json

...

# LangFlow API configuration
LANGFLOW_URL = ""
API_KEY = ""
MCP_COMPONENT_ID = "MCPTools-123" # Insert the component ID of the MCP Connection from LangFlow here.
AUTH_TOKEN = "" # The token you need to pass into MCP.

HEADERS = {
    "Content-Type": "application/json",
    "x-api-key": API_KEY
}

...

@app.route("/send", methods=["POST"])
def send():
    user_message = request.json.get("message")

    payload = {
        "input_value": user_message,
        "output_type": "chat",
        "input_type": "chat",
        "tweaks": { # The key point is!!! This is exactly how header forwarding works from the Langflow API into the MCP Connection
            MCP_COMPONENT_ID: {
                "headers_input": [
                    {
                        "key": "Authorization",
                        "value": AUTH_TOKEN
                    }
                ]
            }
        }
    }

...

[andrewkcarter]

@alexfhub you're the man! Nice work and thank you for the detailed example to work with.

[alexfhub]

It stopped working starting from version 1.5.0. Now MCP is a separate entity in the settings, and unfortunately, it’s no longer accessible via tweaks :(