varunreddy/SkillMesh

A retrieval-gated skill architecture for LLM agents that scales to hundreds of tools by exposing only the top-K relevant capabilities per request.

50
/ 100
Established

Uses BM25 sparse + dense vector retrieval with reciprocal rank fusion to score and select the top-K most relevant tool cards from a registry, then injects only those into the agent prompt. Integrates directly with Claude via MCP server (`uvx` one-liner), Codex skill bundles, and local CLI workflows—all supporting standardized OpenAI function schemas. Includes role-based card grouping (e.g., `Data-Analyst`, `AWS-Engineer`) for multi-domain workflows, reducing prompt bloat from 50K+ tokens (full catalog) to ~3K (retrieval-selected cards).

Available on PyPI.

Maintenance 10 / 25
Adoption 8 / 25
Maturity 18 / 25
Community 14 / 25

How are scores calculated?

Stars

4

Forks

3

Language

Python

License

MIT

Last pushed

Mar 04, 2026

Monthly downloads

118

Commits (30d)

0

Dependencies

5

Get this data via API

curl "https://pt-edge.onrender.com/api/v1/quality/mcp/varunreddy/SkillMesh"

Open to everyone — 100 requests/day, no key needed. Get a free key for 1,000/day.