If you’re tried XMCP — you may have noticed it can search the Xojo documentation. By default that’s keyword-based. This tool upgrades it to semantic search.
I’ve also updated XMCP itself to automatically detect and use the RAG database when it’s present, so the upgrade is seamless once the database is built.
But the database is just a standard SQLite file with embedded chunks of the Xojo documentation — so if you’re building your own tooling around the Xojo docs, it may be useful to you regardless of XMCP.
What it does
XMCP-RAG-Indexer builds a local RAG (Retrieval-Augmented Generation) database from the Xojo documentation. It parses the llms-full.txt file that ships with the Xojo docs and splits it into searchable chunks covering the full API reference and all guide pages. Each chunk is embedded using a local llama.cpp server running the nomic-embed-text model, and stored in a SQLite database that XMCP automatically picks up.
Once in place, XMCP can find relevant documentation even when the exact words don’t match — which makes a real difference when asking the AI about Xojo concepts.
Highlights
- Entirely local — no cloud APIs or subscriptions
- Resumable — interrupted runs continue where they left off
- Idempotent — safe to re-run against the same docs
GitHub: GitHub - o3jvind/XMCP-RAG-Indexer · GitHub
Feedback and contributions welcome!