Receiving an error for the POST /invoke when trying to use as an MCP

Archon
1

i have setup the MCP and start the graph service. It is receiving the prompts from windsurf (Windows version) but it seems it is not able to get to the pydantic ai coder? any thoughts what i missed along the way?

(venv) C:\SOURCE\Archon-main\Archon-main\iterations\v3-mcp-support>python graph_service.py
INFO:     Started server process [19120]
INFO:     Waiting for application startup.
INFO:     Application startup complete.
INFO:     Uvicorn running on http://127.0.0.1:8100 (Press CTRL+C to quit)
20:25:07.543 reasoner run prompt=[Scrubbed due to 'API key']
20:25:07.632   preparing model and tools run_step=1
20:25:07.633   model request
20:25:35.657   handle model response
20:25:35.668 pydantic_ai_coder run stream prompt=[Scrubbed due to 'API key']
20:25:35.695   run node StreamUserPromptNode
20:25:35.703   preparing model and tools run_step=1
20:25:35.704   model request run_step=1
INFO:     127.0.0.1:57689 - "POST /invoke HTTP/1.1" 500 Internal Server Error
20:25:47.603 reasoner run prompt=
    User AI Agent Request: Create a Python script that uses t...o creating this agent for the user in the scope document.

20:25:47.605   preparing model and tools run_step=1
20:25:47.607   model request
20:26:05.003   handle model response
20:26:05.014 pydantic_ai_coder run stream prompt=Create a Python script that uses the NewsAPI.org service to fe... errors gracefully
6. Save the output to a dated markdown file
20:26:05.015   run node StreamUserPromptNode
20:26:05.019   preparing model and tools run_step=1
20:26:05.020   model request run_step=1
INFO:     127.0.0.1:57750 - "POST /invoke HTTP/1.1" 500 Internal Server Error
1 Like
2

That’s strange that there aren’t any useful error logs! I’ll have to look into that since whenever there’s an issue I’d hope it would at least give us something. Are you using local LLMs or which model are you using?

3

image
image1808×703 47.8 KB

Here is my .env. I was using the qwen model for the primary and 03-mini for the reasoner.

1 Like
4

Gotcha! So you unfortunately can’t use OpenAI for the reasoner and a local LLM for the primary model, it has to be both local or both OpenAI/OpenRouter. That’s because the connection to the LLM provider is based on the base URL, and right now you have that set to OpenAI.