Polaris for Cursor
Cursor discovers Polaris through the same .mcp.json that Claude Code uses. No extra configuration — run polaris setup once and Cursor
picks up the MCP server automatically.
How it works with Cursor
Cursor supports MCP server discovery through .mcp.json in your project root. When Polaris is set up, Cursor's AI features can call the polaris.search tool to semantically search your indexed documentation
instead of grepping through files.
This means fewer tokens burned on context-gathering and more accurate answers — Polaris returns ranked, section-aware chunks using hybrid BM25 + vector search with RRF fusion, not raw file contents.
Setup in 30 seconds
curl -fsSL https://raw.githubusercontent.com/girard-g/polaris/main/install.sh | bash
polaris setup # writes .mcp.json — Cursor discovers it
polaris index ./docs # index your markdown
polaris serve # start the MCP serverBackground reindexing
Cursor doesn't support PostToolUse hooks like Claude Code does, so live reindexing
isn't automatic. Instead, run polaris watch in a terminal to watch your indexed
directories for changes. Polaris reindexes modified files in ~500ms, keeping your
search results fresh while you work.
Why not just let Cursor grep?
When Cursor's AI answers a question about your docs, it typically searches with grep and reads matching files — consuming thousands of tokens. Polaris replaces that with a single MCP call that returns only the relevant sections: 10–40× fewer tokens.
See how semantic search compares to grep →