An AI assistant can now read and act on your workspace through the same API the dashboard uses. This release exposes that API as MCP tools and documents both the HTTP surface and the MCP one.
Added
- MCP server — the entire workspace API is exposed as roughly 48 Model Context
Protocol tools, each proxying one API route and handing the JSON response straight
to the model. Two paths reach the same tool table: a local stdio server via
hookwatch mcp, and a remote Streamable HTTP endpoint. Both authenticate with a workspace API key (hwk_…) — no session, no separate credential. - HTTP API documentation — a public API reference with worked
curlexamples, and an MCP page describing the tools, the two transports, and how they authenticate.
Improved
- The API contract was audited against the running server and several drift bugs were corrected, so the documented behaviour matches what the API actually does.
Notes
Release date comes from the v1.17.0 git tag (22 July 2026). The tool table is
identical across the CLI’s stdio server and the remote HTTP server.