The MCP server

What it is

Recited runs an MCP (Model Context Protocol) server at https://app.recitedai.com/api/mcp. Point an MCP-capable AI tool at it with a project API key, and that tool can query your visibility data directly: "which prompts did we lose ground on this week?" becomes a question you ask Claude or Cursor, answered from live project data.

The MCP server exposes the same tool set as the in-app AI agent, so anything the agent can look up, your connected tools can too. The tools cover the visibility overview, brand and competitor performance, prompts and their details, domain and URL citations, gap analysis, recent answers, opportunities, and doc search. With a read-write key, action tools also work: adding prompts and starting article drafts from opportunities.

Setup

Everything you need is on the project's Settings > API keys & MCP page, which shows copy-ready configuration for each supported client with your key filled in. In short:

• Claude Desktop: Settings, Developer, Edit Config, paste the server block into claude_desktop_config.json, restart. Recited shows up under tools.
• Claude Code: add the server, then run /mcp to confirm Recited is connected.
• Cursor: use the one-click install, or Cursor Settings, MCP, Add new MCP server.
• ChatGPT: add Recited under Settings, Connectors.
• Grok: Settings, Connectors, Add MCP server.

The connection itself is a URL plus a bearer key:

{
  "mcpServers": {
    "recited": {
      "url": "https://app.recitedai.com/api/mcp",
      "headers": { "Authorization": "Bearer recitedai_your_project_key" }
    }
  }
}

Exact config shape varies by client; use the in-app snippets, which are always current.

Scope and safety

• The key determines the project; every tool call is scoped to it. Connecting two projects means two server entries with two keys.
• Read-only keys can use every reporting tool. Tools that change data refuse to run without a read-write key, enforced server-side.
• Read tools return trimmed results, the top 25 rows, and date windows default to the last 30 days; ask for a specific date range when you need one.

Usage

Every MCP tool call counts against the project's daily agent and MCP allowance, shared with the in-app agent, plus a 30 per minute cap. See rate limits. If a client starts failing mid-conversation, the allowance is the first thing to check.