INITE Visibility MCP server

INITE MCP — AI visibility, from inside your agent

Ask your assistant whether the answer engines can see a website, and get a number back.

> Can ChatGPT actually find stripe.com, or is it invisible?

  analyze_site(url: "stripe.com")   → run_id: cm4x…
  get_analysis(run_id: "cm4x…")     → Score: 78/100

The audit reads a live site the way an assistant does — the identity files an engine looks for (llms.txt, ai.json and the rest), the crawler policy, the schema graph, page health — and then asks Claude, ChatGPT, Gemini and Perplexity whether they name it. One score out of 100 across eight weighted sections.

It is the same audit that runs at inite.ai/en/analyze.

Works before you sign in

Three of the tools run entirely on your machine — a robots.txt fetched, a file probed for, a page parsed. No account, no allowance, no call home.

> Is anything blocking AI crawlers on stripe.com?

  check_ai_access(url: "stripe.com")

  Retrieval: all 7 answer-engine crawlers may fetch the site.
  Training:  3 of 8 blocked — Meta-ExternalAgent, Bytespider, Amazonbot.

Signing in adds the two that cost something real: the answer engines are actually asked whether they name the site, and the eight weighted sections are scored.

Install

Claude Desktop, Cursor, or any client that launches a stdio server:

{
  "mcpServers": {
    "inite": {
      "command": "npx",
      "args": ["-y", "@inite/visibility"]
    }
  }
}

Then sign in once:

npx @inite/visibility login

That opens a browser, you approve, and the token is stored at ~/.config/inite/mcp.json with owner-only permissions. Authorization code with PKCE over a loopback redirect, the flow RFC 8252 prescribes for a native app. Nothing is written to the repository and no secret ships in the package.

Already have a token (CI, a shared config, a container):

{
  "mcpServers": {
    "inite": {
      "command": "npx",
      "args": ["-y", "@inite/visibility"],
      "env": { "INITE_TOKEN": "…" }
    }
  }
}

INITE_TOKEN wins over the stored file.

Client speaks the MCP authorization flow? Skip this package. Point it straight at https://inite.ai/api/mcp and it will discover the rest: an unauthenticated call answers 401 with WWW-Authenticate naming the protected-resource metadata, which names the authorization server.

Tools

Local — no account:

Remote — needs an account:

Two tools rather than one for the audit, because it is asynchronous. A single tool that blocked for a minute would be torn down by most clients' timeouts.

Commands

inite-visibility            run as an MCP server over stdio (what a client does)
inite-visibility login      sign in through the browser
inite-visibility whoami     say whether a usable token is present

What this package is

Two halves.

The local checks are real work done here: fetching, parsing, and the robots rules applied properly — most-specific group wins, longest matching rule wins, Allow breaks a tie. They cost nobody anything because your machine does them.

Shortened. The full README is on GitHub.