Field notes from the agent web
The WebMCP directory.
Sites, tools and their evidence.
Directory listings
| Site | Status | Category | Platform | Tools | Rank |
|---|
No matching sites.
Try another name or clear the filters.
How we read the evidence.
This is a dated snapshot, not a live monitor. Every production candidate is searchable, including sites where no tools were found. A registration shows that a page exposed a tool; it does not establish that an agent used it.
Two populations, counted separately.
Eight observed states.
The platform effect.
Pages firing Chrome’s WebMCP counter in the mobile crawl. Almost all of the increase came from Shopify; these page counts are separate from the directory’s site counts.
Method. Candidates came from HTTP Archive, Chrome use counters, origin-trial tokens decoded from headers and markup, urlscan, PublicWWW, GitHub, npm and public lists. Each was loaded in Chrome 152 twice: as a normal visitor, then with the trial forced on.
Trial context. At verification, Chrome required origin-trial enrolment to expose native WebMCP. Code, trial tokens and actual registrations are recorded separately. The observed tokens had an expiry of 17 November 2026.
Outside the directory. 249 demo, 140 tooling, 20 dev. SDKs, documentation, starter templates, registries, scanners and hackathon entries were tracked but excluded from the production-site list.
Reading the counts. Site totals describe this inspection set. The timeline counts pages, not sites. The platform-wide Shopify crawl is a separate population; these numbers should not be added together.
Missing a site?
Send it our way. Every submission is checked before inclusion.
