# CyClaw MCP server

> An AI retrieval system that answers questions from local documents using a 12-node LangGraph and local LLMs for privacy-focused inference.

- Repository: https://github.com/cgfixit/CyClaw
- Page: https://hostedmcp.org/servers/cgfixit/cyclaw
- Transport: —
- Stars: 5
- License: MIT

## Environment variables

| Name | Required | Secret | Description |
|---|---|---|---|
| `CYCLAW_API_KEY` | no | no |  |

## FAQ

### How does CyClaw handle document retrieval?

It uses a hybrid search approach combining ChromaDB and BM25 fused through RRF, supported by a 12-node LangGraph.

### Can I use CyClaw offline?

Yes, it supports offline mode where embeddings, BM25, and the cross-encoder run locally on your CPU.

### What happens if the local model fails to answer?

In hybrid mode, it can fall back to external providers like Grok or Claude, but this requires per-question user consent.

### How do I build my document library?

You can run the retrieval indexer command or use the POST /index/build API endpoint.
