Client error '426 Upgrade Required'

[neolee]

I first run the MCP server in streamable HTTP mode using:

uvx mcp-run-python --port 3001 streamable-http

Then test the Pydantic AI client code as below:

from pydantic_ai import Agent
from pydantic_ai.mcp import MCPServerStreamableHTTP

server_http = MCPServerStreamableHTTP(url="http://localhost:3001/mcp")
agent_http = Agent(m.default, toolsets=[server_http])

async def run(agent):
    async with agent.run_mcp_servers():
        result = await agent.run(f"How many days between 2000-01-01 and 2025-12-01?")
    print(result.output)

if __name__ == "__main__":
    import asyncio

    asyncio.run(run(agent_http))

It runs into error:

httpx.HTTPStatusError: Client error '426 Upgrade Required' for url 'http://localhost:3001/mcp'

Anything wrong here?

[DouweM]

@neolee Thanks for the report. What versions of the pydantic-ai and mcp packages are you using?

To get to the bottom of the issue, can you please instrument httpx following the instructions at https://ai.pydantic.dev/logfire/#monitoring-http-requests (logfire.instrument_httpx(capture_all=True)), and then share the complete HTTP request and response traces? Per https://developer.mozilla.org/en-US/docs/Web/HTTP/Reference/Status/426, there should be an Upgrade header detailing what the server wanted from the client.

[neolee]

pydantic-ai-slim[logfire, mcp, openai] v1.28.0 & mcp v1.23.2

logfire output:

{
  "http.method": "POST",
  "http.request.body.text": {
    "method": "initialize",
    "params": {
      "_meta": {
        "traceparent": "00-019b03b0f21dd95c591c7fd8e6ae6661-8a4168afcbbc635a-01"
      },
      "protocolVersion": "2025-11-25",
      "capabilities": {
        "sampling": {}
      },
      "clientInfo": {
        "name": "mcp",
        "version": "0.1.0"
      }
    },
    "jsonrpc": "2.0",
    "id": 0
  },
  "http.request.header.accept": [
    "application/json, text/event-stream"
  ],
  "http.request.header.accept-encoding": [
    "gzip, deflate, br"
  ],
  "http.request.header.connection": [
    "keep-alive"
  ],
  "http.request.header.content-length": [
    "247"
  ],
  "http.request.header.content-type": [
    "application/json"
  ],
  "http.request.header.host": [
    "localhost:3001"
  ],
  "http.request.header.user-agent": [
    "python-httpx/0.28.1"
  ],
  "http.response.header.connection": [
    "keep-alive"
  ],
  "http.response.header.content-length": [
    "16"
  ],
  "http.response.header.content-type": [
    "text/plain"
  ],
  "http.response.header.date": [
    "Tue, 09 Dec 2025 15:18:06 GMT"
  ],
  "http.response.header.keep-alive": [
    "timeout=5"
  ],
  "http.status_code": 426,
  "http.target": "/mcp",
  "http.url": "http://localhost:3001/mcp"
}

p.s. Since we don't have an official streamable HTTP mode sample in the README.md file so I write the code to have a test. But I'm not sure if there are any errors in the code.

[DouweM]

@neolee I can't reproduce the issue I'm afraid; it's working for me with your exact code, command, and package versions, and the request body looks identical to yours.

Can you reproduce this in a clean environment?

[neolee]

@DouweM That's pretty weird. I'll have a look at that later. Thanks anyway 😄