MCP & API
MCP server
What the CiteCue MCP server lets an AI agent read and change, where to find the MCP URL, which plan includes it, and what each role can do through it.
CiteCue runs a remote MCP (Model Context Protocol) server. Connect an AI agent such as Claude, ChatGPT or Cursor to it, and the agent can read your visibility data and, on the projects you can edit, run scans and change prompts, competitors and fix status. You ask in plain language; the agent picks the tools.
A connected agent answers questions like "Which questions did we lose last month?" from the same data the dashboard shows, and handles upkeep such as adding prompts or starting a scan. For worked examples, see MCP use cases.
Where to find it
Open your organization page and choose the API & MCP access tab. From inside a project, open the project selector at the top of the sidebar and choose All projects, use API & MCP access in the Organization card on the project's Settings page, or open Integrations and click Connect an agent on the Connect an AI agent (MCP) card.
The Connect an AI agent card shows the MCP URL, with a copy button and setup steps for each supported client:
https://app.citecue.com/api/mcp
The same page lists your Connected agents and, for organization owners, the organization's API keys.
Which plan includes it
MCP access is included in Optimize, Agency and the active seven-day Optimize trial. The Connect an AI agent card is labelled Optimize plan. Compare plans on the pricing page. On Measure or another plan without MCP access, the card shows a preview with Upgrade to Optimize instead of the URL and setup steps.
An agent can still connect on any plan, and list_projects lists your projects with mcpAccess set to locked. Every other tool returns a plan_locked error saying MCP access isn't included in the organization's plan, with an upgradeUrl to your organization's billing page.
Two ways to connect
- Claude.ai, Claude Desktop and ChatGPT sign in: you add the MCP URL as a connector, sign in to CiteCue and click Allow. Follow the sign-in steps for Claude and ChatGPT.
- Claude Code, Cursor and Codex CLI send an organization API key that an owner creates. Follow the API key setup for Claude Code, Cursor and Codex CLI.
Who can do what
| Who | Read data | Change scans, prompts, competitors and fix status |
|---|---|---|
| Project Member | Yes, on that project | No |
| Project Admin | Yes, on that project | Yes, on that project |
| Organization Owner | Yes, every project | Yes, every project |
| API key | Yes, every project in its organization | Yes, every project in its organization |
A connector you sign into acts as you. It reaches every organization you belong to, sees exactly the projects you can open in the dashboard, and has your role on each one. A Member's agent can read but gets a forbidden error when it tries to change something.
An API key acts as an Owner of the organization that created it and sees nothing outside that organization.
What it can do, and what it never can
The server has 23 tools. The MCP tools reference lists every parameter and result.
Reads, on any project you can open: visibility and the latest report, competitors by topic, prompts, cited domains, lost questions, factor gaps, insights and the fix queue, sentiment, AI referrals, AI Readiness, Agent Usability, and your projects and scans.
Changes, on projects where you are an Admin or the organization Owner:
- start a scan and check on it, which counts against your monthly manual-scan allowance like Run scan;
- add prompts, edit one, or switch one off;
- track a new competitor, one CiteCue suggested or one you name, which only the dashboard can remove again;
- move a fix between New, Planned, In progress and Resolved, or set it aside from recommendations without closing it.
It can never:
- delete anything. There is no delete tool, so a prompt is switched off rather than removed and keeps its history;
- touch billing, plans or payment methods;
- invite, remove or change the role of anyone on your team;
- see an organization or project the account behind it can't;
- change a project where your role is Member.
Nothing an agent does is hidden
An agent's changes show in the dashboard like anyone else's. A prompt it adds appears on the Prompts page, with the source User. A scan it starts is a manual scan: it counts against your monthly allowance and you can stop it from the dashboard with Stop scan.
To cut an agent off, revoke its connection under Connected agents or revoke its API key; its next request fails. For the product overview, see the CiteCue MCP server page.