WebMCP page tools

Three Busymate websites now publish tools an AI agent can call directly in the browser, using the WebMCP ModelContext API. When you open one of these pages in an agentic browser, the page hands the agent a small set of read-only functions instead of making it scrape the HTML.

Every tool is read-only, answers from the same data the page renders, and reaches nothing the page could not already show you.

What each site offers

Documentation — `docs.busymate.dev`

Tool	Ask it for
`search_docs`	Pages matching a question or keywords, each with a snippet. Takes `query`, optionally `limit`.
`read_docs_page`	One page's full markdown. Takes `slug`, e.g. `quickstart` or `dashboard/feed`.
`list_docs_sections`	The table of contents — every section and its pages, in sidebar order.

They read the same _meta.json registry and markdown files this site renders, so an answer is never a stale copy of a page. Internal pages are invisible to them for the same reason they 404 here.

Marketing — `busymate.dev`

Tool	Ask it for
`get_product_overview`	What Busymate DevTools is, every surface, how capture works, and the honest "what it is not" list.
`get_pricing`	The tiers, the feature comparison and the FAQs — flagged as a preview, not purchasable, exactly as the page says.
`get_install_command`	The install command and links for one client. Takes `client`: `bmc`, `mcp`, `ios` or `android`.
`get_download_links`	Every place to get the product: App Store, Google Play, the APK, the bmc script, the MCP endpoint.

Status — `status.busymate.dev`

Tool	Ask it for
`get_status_summary`	The live snapshot: overall verdict, services responding, host load and uptime, upstream health, build drift.
`get_service_status`	One service in full. Takes `service` — a key or a label, e.g. `proxy-server` or `Dashboard`.

Both read the board's own /api/status snapshot, so a tool answer and the card on screen are the same numbers from the same read.

What an agent should expect

Read-only. Every tool is annotated readOnlyHint: true. None of them changes anything, signs you in, or writes data.
No hidden reach. A tool sees exactly what the page already serves publicly. There is no credential, no token, and no private endpoint behind any of them.
Honest answers. get_pricing says in its own response that the figures are a preview awaiting sign-off; get_service_status refuses an unknown service name rather than returning a plausible neighbour.

Browser support

WebMCP is new. Where the browser exposes document.modelContext — Chrome and Edge with the origin trial enabled, or with chrome://flags/#enable-webmcp-testing turned on — the tools appear to the agent automatically. Where it does not, nothing is broken: the page renders exactly as before and the tools simply are not offered.

These are page tools, distinct from the MCP server at mcp.busymate.dev. The MCP server is your authenticated, full-power surface — hundreds of tools that read and change your own capture data. The page tools here are anonymous, read-only, and scoped to what one public website already shows.