Implement the WebMCP browser API so an in-browser AI agent can invoke actions on the docs site. WebMCP is client-side JS (navigator.modelContext.provideContext()) — implementable in this repo via Docusaurus theme JS, no backend/infra.
Note: WebMCP ≠ an MCP server. This is per-page browser tooling, not a network endpoint.
Scope
A docs site is read-heavy (agents already read our content via llms.txt / api-catalog / markdown / agent-skills), so the one genuinely useful action to expose is:
search_docs(query) — search the docs, return matching pages/snippets. Backed by our existing search index.
Each tool needs name, description, inputSchema (JSON Schema), and an execute callback.
The only caveat is that this is experimental. WebMCP is Chrome EPP + W3C draft, limited support today.
Refs
Implement the WebMCP browser API so an in-browser AI agent can invoke actions on the docs site. WebMCP is client-side JS (
navigator.modelContext.provideContext()) — implementable in this repo via Docusaurus theme JS, no backend/infra.Scope
A docs site is read-heavy (agents already read our content via llms.txt / api-catalog / markdown / agent-skills), so the one genuinely useful action to expose is:
search_docs(query)— search the docs, return matching pages/snippets. Backed by our existing search index.Each tool needs
name,description,inputSchema(JSON Schema), and anexecutecallback.The only caveat is that this is experimental. WebMCP is Chrome EPP + W3C draft, limited support today.
Refs