Build against preset AI.
Technical reference for the surfaces AI tools and scripts talk to: the MCP server, its tools, and the command line. If you are looking for guides and concepts instead, those live in the documentation.
Where to start
Read the protocol pages first. The per-tool and per-command references assume you know how tools are categorised and when a write actually lands.
The three layers, what each one is responsible for, and the usage pattern most integrations should follow.
Five categories, how to choose between them, and how behaviour changes with the active mode.
Which tools only observe, which ones propose changes, and why the split matters for agents.
Propose, validate, apply. What each step guarantees and where a change can still be rejected.
Reference
Generated from the source in packages/mcp and packages/cli rather than hand-maintained alongside it.
Per-tool pages with parameters, types, and return shapes, generated from the server's own schemas.
Usage, arguments, and options per command, including scan, validate and ci.
Still being written. Framework-aware quickstarts, the HTTP API reference, and published schemas are in progress. Until they land, the enforcement guide covers per-tool setup for Cursor, Windsurf, VS Code Agent, and Zed.