Connect your AI to CiteHawk.
Build on your AI visibility data. Every score, ranking, gap, and real AI answer CiteHawk collects is available over a REST API, an MCP server, and a Claude Code skill, so your own agents can read your visibility and act on it.
Ways to connect
New to the API? Start with the API quickstart, which gets you from zero to your first call.
REST API
Full reference for the /api/v1 endpoints: brands, metrics, responses, and scores. JSON over HTTPS, Bearer auth.
View the referenceModel Context ProtocolMCP server
Connect Claude, Cursor, or any MCP client and ask your AI about your visibility. Six read-only tools.
Connect a clientAgentic GEOClaude Code skill
The find-fix-measure loop: Claude Code reads your gaps via MCP, edits your own site repo, and the next collection measures the lift.
Install the skillCustom GPTs + agentsChatGPT & other agents
Use the OpenAPI spec to wire CiteHawk into a ChatGPT custom GPT Action or any agent framework that speaks OpenAPI.
Wire up an ActionOne-click connector
The MCP server supports the OAuth 2.1 connector flow, so in a client that speaks it, such as Claude, you add CiteHawk by URL and approve it in the browser. There is no key to create, copy, or paste.
The connector URL is:
https://www.citehawk.com/api/mcpWhat happens
- You add the URL as a custom connector in your client. The client discovers that CiteHawk requires authorization and registers itself automatically.
- Your browser opens a CiteHawk consent screen. If you are not signed in, you sign in first and land back on it.
- The consent screen names the client and shows the host your data will be sent to. Read that host. Anyone can register a connector and choose its name, so the destination is the part worth checking, and CiteHawk says plainly when it does not recognize one.
- You choose one workspace to grant access to, then approve or deny. Access is read-only and covers a single workspace.
- The client receives a token and can read your visibility data. Tokens are short-lived and refresh in the background.
A workspace that has never run a collection has nothing for an AI model to read, so the consent screen asks you to start your first report before connecting.
Pasting an API key
Clients that do not implement the OAuth connector flow authenticate with a workspace API key instead, sent as a Bearer token. See API keys for how to create one and the MCP server page for per-client configuration.
One restriction matters here. The MCP server accepts an unscoped API key only. Its tools span scores, metrics, rankings, source gaps, recommendations, and evidence, which is broader than any single REST scope, so accepting a key that its creator deliberately restricted would quietly undo that restriction.
A scoped key is rejected as unauthorized without a specific explanation, so this is worth knowing before you debug it. If an MCP client reports an authorization failure and your key works against the REST API, check whether the key has scopes selected. Use an unscoped key, or connect over OAuth.
Which should I use?
| If you are… | Use |
|---|---|
| Adding CiteHawk to a client that supports connectors | The one-click OAuth flow |
| Configuring an MCP client by hand or in a config file | An unscoped API key |
| Calling the REST API from a script or backend | An API key, scoped to what the script needs |
Revoking access
Connections made over OAuth are listed under Connected apps in your account settings, with the client, the destination host, the workspace it can read, and when it was last used. Revoking there cuts the connection off immediately.
API keys are revoked separately, from the workspace settings page. See API keys.
