Quick start

Pinax is a single Go binary. Install it, point it at a docs site, then run pinax serve. Any MCP-compatible client - Claude Desktop, Claude Code, Cursor, Windsurf, Cline - can connect over stdio. No daemon, no cloud, no re-indexing pipeline.

1. Install

The fastest path on macOS or Linux:

brew install desmondsanctity/tap/pinax

If you prefer Go:

go install github.com/desmondsanctity/pinax/cmd/pinax@latest

Or grab a pre-built binary from the releases page and drop it on your $PATH. The full matrix of install options lives in Configuration.

2. Index a docs site

Pick a name from the catalog:

pinax add stripe

Or point at any URL - anything containing ://, a dot, or a slash is treated as a URL:

pinax add https://docs.convex.dev

Pinax discovers pages via llms.txt, then sitemap, then a bounded BFS crawl, and writes an atomic JSON manifest plus a BM25 index under ~/.pinax/servers/<name>/.

3. List what’s indexed

pinax list

You’ll see every manifest with its page count, base URL, and last-refresh timestamp.

4. Serve over stdio

The same command works for every MCP-compatible client:

pinax serve

That serves every indexed manifest at once. To pin a single one:

pinax serve stripe

Stdio is what MCP clients launch on demand - you usually don’t run this in a terminal, you point your client config at it. See Serve for the HTTP transport + log UI.

5. Connect a client

Pick yours: Claude Desktop · Claude Code · Cursor · Windsurf · Cline.

Or have Pinax print the snippet for you:

pinax config claude --project

What you get

Four tools, scoped to one manifest or every manifest at once: list_sections, search_pages, get_section_pages, get_page. Plus list_docs to enumerate what’s loaded. See MCP tools for arg shapes and return contracts.

Where to next

• Catalog - what pinax add <name> resolves to and how to refresh.
• Serve - stdio vs HTTP and the log viewer.
• Troubleshooting - JS-rendered sites, doctor output, common errors.