MCP server for Aim. Exposes runs, params, and metric data to LLMs.
I use this to ask the following questions:
- Analyze the metrics from the most recent run tagged with 'flux' and summarize how the changes appear to be working.
- Compare gradients between runs
6d6c3226...ande2ea06cc..., focusing on AdaLN weight gradients, and explain the key takeaways. - Review the latest run
c37d034b...and interpret thex,y, andzmetrics.
python -m venv venv
source venv/bin/activate
pip install -r requirements.txt
python aim_mcp.py --repo /path/to/aim/repo --transport sse --port 8000
- Minimal work went into this, built over a weekend with Gemini doing most of the coding. However, the codebase is small
- Aim appears to be marginally maintained
- There are no tests
- Best-effort help and support only