First-class for AI coding agents.
Cuitty's audience is humans and the agents they delegate to. Every doc is also a JSON object and a Markdown blob. Every example uses real, signed events against the real wire protocol.
The discovery file. A short, ordered index of every doc page in the corpus.
Every Markdown source, concatenated, with `<!-- path: -->` separators. One fetch and your agent has the whole knowledge base.
Structured doc index — slug, title, headings, frontmatter, links. Pagefind-compatible. Use it for tool-augmented retrieval.
Read-only MCP server: search_docs, get_doc, list_modules, wire_protocol_example. Manifest at /.well-known/mcp.json.
Wire-protocol first
Wire-protocol first.
Cuitty isn't an SDK with an HTTP escape hatch. It's an HTTP contract with SDKs as conveniences. Anything an SDK can do, an agent can do with curl.
Every doc has a JSON twin.
For `/docs/foo` you also get `/docs/foo.md` (raw Markdown) and `/docs/foo.json` (structured). Phase 2 ships these route handlers.
Robots are welcome.
robots.txt explicitly allows GPTBot, ClaudeBot, PerplexityBot, Google-Extended, and cohere-ai. We disallow only ephemeral search indexes.
MCP server on the way.
An HTTP-based MCP endpoint at /mcp will expose read-only tools — `search_docs`, `get_doc`, `list_modules`, `wire_protocol_example`. Manifest documented today at /.well-known/mcp.json — see /agents/mcp.
If you're a human reading this, the same surfaces work great for you — paste them into Claude, Cursor, or Aider and the model has the full Cuitty corpus.