Signpost.

One objective. Many sites.

Signpost points WebMCP agents to the surfaces that can do what they need. The agent decides what comes next.

It points. It does not drive.

Declared capabilities

provider-declared · click a row for the response

GET /resolve?capability=order-coffee        200 · 9ms
{ "capability": "order-coffee",
  "surfaces": [ { "origin": "https://valentincoffee.dev",
                  "webmcp": "/webmcp",
                  "tools": ["menu.browse", "order.create"] } ] }
# stateless — no objective, no session, no next step

GET /resolve?capability=book-haircut        200 · 7ms
{ "capability": "book-haircut",
  "surfaces": [ { "origin": "https://timothygeorge.studio",
                  "webmcp": "/webmcp",
                  "tools": ["availability.list", "appointment.book"] } ] }

GET /resolve?capability=check-hex-code      200 · 8ms
{ "capability": "check-hex-code",
  "surfaces": [ { "origin": "https://sirocco.gallery",
                  "webmcp": "/webmcp",
                  "tools": ["hex.check", "conformance.report"] } ] }

Recent resolutions

09:14:02order-coffee → valentincoffee.dev9ms

09:14:31book-haircut → timothygeorge.studio7ms

09:15:06check-hex-code → sirocco.gallery8ms

09:15:44deliver-parcel → no surface declares this4ms

The boundary

Providers declare capabilities. Signpost resolves one at a time. The agent does everything else.

Each resolution is independent and stateless. Signpost never receives the objective, never holds a session, and never tells an agent what to do next. Providers keep their own WebMCP tools; the agent calls them directly on their own sites.