An MCP server for AI brand visibility

What you get

Four tools, each a thin pass-through to a REST endpoint. The MCP server is free to install. Each tool call is billed at the underlying endpoint’s credit rate.

mentions_check — check a brand against a prompt. Default path is mode:quick (ChatGPT, Claude, Gemini, and Perplexity via their APIs): 150 credits fresh, 5 on a cache hit. If you ask for the live Perplexity UI, the tool uses mode:perplexity_live at 85 credits. The mode is picked from the natural-language ask.

mentions_watch — create a scheduled brand-rank watch with webhook delivery from inside the conversation (POST /v1/watch).

mentions_discover — generate query candidates for a brand without leaving the editor (POST /v1/discover, 165 credits).

mentions_compare — a head-to-head delta between two brands or two queries (POST /v1/compare, 500 credits).

Setup guides: Claude Code, Cursor, Windsurf. ChatGPT Plus is different: it uses a Custom GPT Action and an OpenAPI paste, not MCP. That walkthrough is here.

Remote server — OAuth, no install

MentionsAPI hosts a remote MCP server with OAuth, so claude.ai and Claude Desktop can connect without installing a package or pasting a key.

https://api.mentionsapi.com/mcp

On claude.ai: Settings → Connectors → Add custom connector. Paste the URL, click Connect, sign in to MentionsAPI if prompted, and click Allow. Connecting creates an API key named “Claude” under /app/keys. Revoke that key to disconnect.

The same URL works in any Streamable-HTTP MCP client with an Authorization: Bearer lvk_live_… header. Ready-made snippets are in the dashboard.

Local server — npm package

One local process, your key, no OAuth flow. Replace lvk_live_... with a key from /app/keys.

npx -y @mentionsapi/mcp@latest --api-key=lvk_live_...

The package is a thin wrapper: it forwards the bearer token and turns MCP JSON-RPC into HTTPS calls. Nothing is processed locally. The API key is the only secret.

Cursor (.cursor/mcp.json in the project, or ~/.cursor/mcp.json globally):

{
  "mcpServers": {
    "mentionsapi": {
      "command": "npx",
      "args": ["-y", "@mentionsapi/mcp@latest"],
      "env": {
        "MENTIONSAPI_API_KEY": "lvk_live_..."
      }
    }
  }
}

After a restart, ask the composer: “Use the mentionsapi tool to check if Notion is mentioned in ‘best note-taking app’.” It should call mentions_check.

Claude Code and Windsurf use the same command / args / env shape. Exact file paths are in the Claude Code and Windsurf guides.

The same API, if you would rather call HTTP

MCP is convenient when the caller is an agent. When the caller is your code, skip MCP and POST /v1/check yourself. Authentication is a bearer token. The body is mode, query, and brand — the same fields the check tool sends.

This is the request on the live quickstart and the /v1/check reference:

curl https://api.mentionsapi.com/v1/check \
  -H "Authorization: Bearer $MENTIONSAPI_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "mode": "quick",
    "query": "best CRM for small business",
    "brand": "HubSpot"
  }'

The check reference returns one object per engine, with mentioned, rank, context, and citations. A shortened mode:quick example from that page:

{
  "request_id": "req_01JBQ2K8H...",
  "id": "chk_a8d2c...",
  "mode": "quick",
  "query": "best CRM for small business",
  "brand": "HubSpot",
  "providers": {
    "chatgpt": {
      "mentioned": true,
      "rank": 6,
      "context": "Top picks include HubSpot CRM...",
      "citations": [
        {
          "url": "https://zapier.com/blog/best-crms/",
          "title": "Zapier",
          "text": "HubSpot CRM stands out for its free tier"
        }
      ],
      "fan_out": []
    },
    "claude":  { "mentioned": true, "rank": 4, "context": "...", "citations": [] },
    "gemini":  { "mentioned": true, "rank": 7, "context": "...", "citations": [] },
    "perplexity": { "mentioned": true, "rank": 6, "context": "...", "citations": [] }
  },
  "cache_hit": false
}

Live UI surfaces use the same endpoint with a different mode: ai_overview, ai_mode, bing_copilot, chatgpt_live, gemini_live, perplexity_live, or all_live. See the mode reference and the methodology page for how mentions are extracted.

What a check costs

The MCP server does not have a separate price. Signup includes 300 free credits and does not require a card. After that, monthly is $5 for 1,000 credits; annual is $54 for 12,000 credits granted up front. Paid plans can buy top-up packs ($4.00 per 1,000 credits on monthly, $3.00 per 1,000 on annual).

Posted rates that the MCP tools actually hit:

Brand extraction is included. If every provider in a call fails, that call is free. Full table: pricing. Usage: GET /v1/usage.

When this fits (and when it doesn’t)

This fits if you write code or already work in Claude Code, Cursor, or Windsurf, and you want a structured mention check inside that workflow — or inside a product you are shipping. It also fits if you need a live check for an arbitrary prompt and brand, not a report pulled from a pre-built monitoring project.

It is the wrong purchase if you want a marketing dashboard, a weekly PDF, or a vendor who will interpret the numbers for you. In that case a managed AI-visibility platform is the honest recommendation. If you are comparing MentionsAPI to a dashboard product, start with MentionsAPI vs Peec AI or MentionsAPI vs Profound.

The /docs/mcp pages remain the canonical setup instructions. This page is the product explanation; those pages are the click-by-click config.