Agent ReadySign in

Agent Ready, in your MCP client

Run agent-readability scans from Claude Desktop, Claude Code, Cursor, Cline, VS Code, or Windsurf. Same scanner, same key, three install paths.

Last updated

Model Context Protocol (MCP)

Drop Agent Ready into Claude Desktop, Claude Code, Cursor, Cline, VS Code, or Windsurf to run scans inline and query history. Three install paths — pick whichever your client supports best. All three use the same Pro API key.

Claude Desktop and Cursor support remote HTTP MCP servers natively. Paste this into your client’s config and replace the key with one from your dashboard.

{
  "mcpServers": {
    "agent-ready": {
      "url": "https://agent-ready.dev/api/v1/mcp",
      "headers": {
        "Authorization": "Bearer ar_live_..."
      }
    }
  }
}

Tools: scan_site(url, pageLimit?), get_scan(id). Prompts: scan, interpret_scan, remediation_plan. Resources: methodology, checks, llms.txt, specs. The hosted server manifest is published at /.well-known/mcp.json.

Where is Agent Ready listed?

The MCP server is published across the major MCP registries and marketplaces. Each link below opens the canonical listing for the channel.

Smithery →

Hosted MCP server registry and proxy. Includes the .mcpb bundle for local installs.

MCP registry →

Official Model Context Protocol registry. Namespace: io.github.mlava/agent-ready-mcp.

npm →

Published as agent-ready-mcp. Installed via npx on demand — no global install needed.

GitHub →

Source, issue tracker, MIT licence, and v0.1.1 release with the .mcpb bundle attached.

mcp.so →

Community MCP server directory. Listing page surfaces the install snippet and tool/prompt rundown.

What does the MCP server expose?

Two tools, three prompts, one Claude Skill descriptor. All four are advertised in the discovery manifest at /.well-known/mcp.json and surfaced in any MCP client’s tools/prompts panels.

KindNamePurpose
toolscan_siteRun a fresh scan against a URL.
toolget_scanFetch a previous scan by id.
promptscanFresh scan + high-level summary.
promptinterpret_scanPlain-English explanation of a scan’s findings.
promptremediation_planPrioritised fix-it doc (optional focus: “seo” or “agents”).
resourcemethodologyScoring formula, rating bands, JS-rendering handling.
resourcechecksReference table of all 59 check IDs.
resourcellms.txtAgent Ready’s own /llms.txt mirrored.
resourcespecsCanonical URLs for the seven specs we validate.
skillagent-readyAnthropic Claude Skill descriptor for activation routing.

How do the three install paths differ?

All three call the same scanner with the same API key. The choice comes down to which transport your MCP client supports best.

  • Remote (HTTP) — point the client at https://agent-ready.dev/api/v1/mcp with a Bearer header. Lowest-latency option for clients that speak Streamable HTTP MCP natively (Claude Desktop, Cursor, the Vercel MCP consumer).
  • Local (npm) — spawn npx -y agent-ready-mcp@latest as a stdio child process. Same key delivered via AGENT_READY_API_KEY env var. The npm package is a thin stdio wrapper that forwards to the hosted REST API. Use this for stdio-only clients or when you don’t want to lean on the community mcp-remote bridge.
  • mcp-remote bridge— community shim that exposes a remote HTTP MCP server as a local stdio process. Use this if your client only speaks stdio and you don’t want to install a per-server package.

Frequently asked questions

What's the difference between the hosted endpoint and the npm package?

Both call the same scanner and require the same Pro API key. The hosted endpoint at agent-ready.dev/api/v1/mcp speaks Streamable HTTP — Claude Desktop and Cursor connect to it directly. The npm package (agent-ready-mcp) is a stdio wrapper that any MCP client can spawn locally; it forwards requests to the same hosted endpoint over HTTPS. Pick HTTP for native clients, npm for stdio-only clients (some IDE plugins, Cline) or when you want zero round-trip via the community mcp-remote bridge.

Which MCP clients can I use this with?

Claude Desktop, Claude Code, Cursor, VS Code (with the MCP extension), Cline, Windsurf, and any other client that speaks the Model Context Protocol. The hosted HTTP endpoint covers clients that support remote MCP servers; the npm package covers stdio-only clients via npx.

What tools, prompts, and resources does the server expose?

Two tools — scan_site (run a fresh agent-readability scan against a URL) and get_scan (fetch a previous scan by id). Three prompts — scan, interpret_scan, remediation_plan — that wire the tools into end-to-end workflows. Four resources — methodology (the scoring formula), checks (table of all 59 check IDs), llms.txt (Agent Ready's own /llms.txt), and specs (canonical URLs for the seven specs we validate). One Claude Skill descriptor (skills/agent-ready/SKILL.md inside the npm package) for activation routing.

Do I need a paid plan?

Yes. The MCP server uses the same Bearer-token authentication as the REST API and only accepts Pro or Team keys. Free-tier scans are available through the web UI but not via the MCP server. Issue a key from your dashboard after upgrading.

Where can I find the source and report issues?

GitHub at github.com/mlava/agent-ready-mcp. The repo includes the full source, test suite, build pipeline, manifest.json (Glama Marketplace), server.json (MCP registry), and glama.json (Glama maintainer claim). MIT-licensed.

REST API & CI/CD →

Same scanner via REST endpoints or the GitHub Action.

Scoring methodology →

How the 59 checks combine into the 0–100 score.