AshutoshRaj97/agentready-mcp

What is AgentReady?
AgentReady is a hosted capability layer that makes any website discoverable and usable by agents through MCP.
Paste a URL → AgentReady crawls the site, generates a spec-compliant llms.txt, and hosts a live /ask RAG endpoint and MCP server. Any MCP-compatible agent client can then discover the site, ask natural-language questions, inspect capabilities, and create grounded read-only plans with citations.
**The problem it solves:** AI agents using web_fetch fetch one page at a time, get empty HTML from JavaScript SPAs (React, Next.js, Vue), and hallucinate when the answer spans multiple pages. AgentReady indexes the whole site, handles JS rendering, and retrieves across pages — so agents get the right answer instead of a confident wrong one.
**What's already indexed:** Browse the live AgentReady directory → or index any public site yourself.
**Key properties:**
- Works on any public URL — static sites, React/Next.js SPAs, Docusaurus, GitBook, custom engines
- No account required to index your first site
- Shared index — one team member submits a site, everyone on the team can query it instantly
- Handles JS-rendered pages that web_fetch returns empty for
- Capability manifests expose freshness, schemas, endpoints, and read-only limits
- Grounded plans return evidence, risks, confirmation requirements, and durable receipts
CLI
The same package doubles as a CLI — no install, no account:
For CI, dashboards, or scripts, add --json to receive the raw report on
stdout (progress remains on stderr):
The command still exits 1 when the grade is below B.
grade exits non-zero below a B, so you can use it as a CI gate. refresh in your docs deploy pipeline keeps the index fresh automatically:
---
Connect any MCP-compatible client to AgentReady:
Local MCP bridge (stdio)
For clients that use an mcpServers configuration, add the AgentReady bridge:
Restart your client. You'll have eight tools available:
- list_sites — see all indexed websites
- get_site_capabilities — index on demand, then inspect a site manifest, freshness, schemas, and endpoints
- get_site_facts — key structured facts extracted from a site
- ask_site — query any site with cited, multi-page answers
- plan_site_action — index on demand, then create a grounded, read-only plan and receipt
- submit_site — index any website so it can be queried
- refresh_site — re-crawl a site, or perform its initial index when it is new
- rate_answer — submit quality feedback
Cursor
Add to ~/.cursor/mcp.json:
Project or terminal setup
Or add the same agentready server to your client's project configuration to share it with your team:
VS Code (GitHub Copilot agent mode)
Requires VS Code 1.99+ with the GitHub Copilot extension. Create .vscode/mcp.json in your project root:
Copilot's MCP tools are only available in agent mode. Commit this file to share with your team.
Windsurf / Zed / other clients
Any MCP client that supports stdio transport works the same way — use npx -y @agentreadyweb/mcp as the command.
WebMCP (no install)
If your client supports HTTP transport, connect directly by URL — no npm required:
Docusaurus plugin
If your docs site uses Docusaurus, auto-index on every build:
MkDocs plugin
If your docs site uses MkDocs, auto-index on every build:
If site_url is not set, specify the domain explicitly: