An agent using a plain fetch reads one page at a time, gets empty HTML from a React site, and answers confidently anyway. Indexing the whole site and retrieving across it is the difference — and plan_site_action is deliberately read-only, so a grounded plan never quietly becomes an action nobody approved.
A hosted capability layer that makes a website usable by agents. Paste a URL, it crawls the site, generates a spec-compliant llms.txt, and hosts a live retrieval endpoint — so a question returns a cited answer synthesised across pages, including from JavaScript-rendered sites a plain fetch returns empty for.
- ask_site answers a natural-language question about a site with citations, across multiple pages
- submit_site indexes a site through a four-layer pipeline that handles React, Next.js and Vue single-page apps
- list_sites shows what is already indexed — the index is shared, so a colleague's submission is queryable immediately
- get_site_capabilities returns a site's manifest — freshness, read-only limits, schemas, and available MCP and HTTP endpoints — indexing the site first if it is new
- plan_site_action turns a request into a grounded read-only plan with sources, risks and confirmation requirements, and never executes side effects
- refresh_site forces a full re-crawl when content has changed
- rate_answer submits quality feedback tied to a specific earlier answer
- The same package doubles as a CLI: grade produces an agent-readiness report card, plus ask, index and refresh, with --json for CI and a non-zero exit below a B
The npm package @agentreadyweb/mcp over stdio with npx -y, or the hosted endpoint https://www.agentready.it.com/api/mcp for clients that speak HTTP. No account is required to index your first site. AGENTREADY_MCP_URL overrides the endpoint for a self-hosted instance. Build plugins exist for Docusaurus, MkDocs and Starlight so a docs site re-indexes on every deploy.
One command — npx @agentreadyweb/mcp