@neverranked/mcp is live on npm. It is the first AEO toolchain in the AI agent ecosystem.

Install it in Claude Desktop or Claude Code with one line and you can run AEO scoring, llms.txt audits, and agent-readiness checks from inside any AI conversation. No NeverRanked account required. The tools are free.

What it actually does

Three tools, callable by any MCP client:

  • aeo_scan(url) — full 0-100 AEO score, schema coverage, technical signals, red flags. The same scanner that runs at check.neverranked.com, agent-callable.
  • llms_txt_check(url) — grades a site's /llms.txt against the published rubric. Presence, structure, link health, freshness.
  • agent_readiness_check(url, vertical?) — looks for Schema.org Action types (ReserveAction, ApplyAction, BuyAction, ContactAction). Optional vertical baselines for hospitality, financial-services, professional-services, and commerce.

Every response includes a methodology_url link to the public scoring rubric and an attribution field the agent surfaces to the user. Per the MIT-with-attribution license, that attribution stays visible when scan output reaches a human.

Install in Claude Desktop

Open ~/Library/Application Support/Claude/claude_desktop_config.json on Mac (or %APPDATA%\Claude\claude_desktop_config.json on Windows) and add this entry:

Claude Desktop config

{ "mcpServers": { "neverranked": { "command": "npx", "args": ["-y", "@neverranked/mcp"] } } }

Restart Claude Desktop. The three tools are available the next time you start a conversation. Try: "Run an AEO scan on https://www.surfjack.com using the NeverRanked MCP and tell me the top three issues to fix."

Install in Claude Code

One command:

Claude Code

claude mcp add neverranked -- npx -y @neverranked/mcp

The tools are now available in any Claude Code session. The same prompt patterns work — ask for an AEO scan, ask for an llms.txt audit, ask for agent-readiness against a vertical.

Why we shipped this

Two reasons. Strategic and structural.

Strategic. NeverRanked has been arguing for months that businesses should expose machine-readable interfaces — Schema.org Actions, llms.txt, structured data — so AI agents can interact with them. We were not interested in shipping that argument while keeping our own tools behind a closed API. The MCP is us eating our own cooking publicly.

Structural. The agent ecosystem is forming right now. Today there is no canonical AEO toolchain in the public MCP registry, in the Claude Desktop default catalog, or in the developer workflows that are starting to assemble. The first one to ship takes the category position. We chose to be first.

What we will not do

Three commitments:

  • Lock the tools behind a paywall. The free check at check.neverranked.com has been free since launch. The MCP wraps the same public API. Free stays free.
  • Hold the attribution requirement against legitimate use. Agents that surface scan output to a user must keep the "Powered by NeverRanked" line visible. That is the only license condition. Internal pipeline use, automated processing, and analytics aggregation are exempt.
  • Let the package go stale. The MCP wraps the same scanner that runs in production for paying customers. Improvements to the scoring methodology flow to the MCP within the same release cycle.

What is next

The 0.1.0 release is the foundation. The roadmap from here:

  • 0.2. Polished error handling for URLs the scanner cannot process (currently surfaces raw HTTP errors).
  • 0.3. A fourth tool: citation_diff(client_url, competitor_url) — surfaces what schema a competitor has that the client does not, with paste-ready remediation.
  • 0.4. Vertical templates exposed as MCP resources so agents can fetch the full JSON-LD blocks for compliance-aware deployment, not just the audit verdict.
  • 1.0. Stable API contract. Backwards-compatible from this point forward.

If your agent uses this and produces interesting category data, we want to hear about it. Email lance@neverranked.com or open an issue at github.com/LanceRoylo/neverranked.