MCP server

The Model Context Protocol (MCP) is an open standard that lets AI clients call external tools. CiteHawk ships an MCP server so you can connect your own AI and ask it about your visibility in plain English. It reads your live data, you never export a thing.

Endpoint

The server speaks streamable HTTP at:

https://www.citehawk.com/api/mcp

Every tool is read-only and scoped to a single workspace. Tools never take a workspace argument: the workspace is fixed by the credential, so a connected AI model cannot reach across your workspaces.

Authentication

There are two ways in. Clients that implement the MCP authorization flow can connect with no key at all; everything else uses a workspace API key.

One-click connector (OAuth)

CiteHawk runs an OAuth 2.1 authorization server alongside the MCP endpoint. In a client that supports it, such as claude.ai, you add the endpoint URL as a custom connector and approve it in the browser. The client registers itself, you sign in, pick one workspace, and grant read-only access. No key is created, copied, or stored in a config file.

The consent screen shows the host your data will be sent to. That is the part worth reading: anyone can register a connector and name it whatever they like, so the destination is what tells you who you are actually connecting.

Access tokens are short-lived and refresh automatically. Revoke a connection at any time under Connected apps in your account settings. See Connect your AI for the step-by-step.

API key

Otherwise, authenticate with the same workspace API key used everywhere else, passed as a Bearer token. Generate one in Settings > API Keys, or see API keys.

The key must be unscoped. These tools read across scores, metrics, rankings, source gaps, recommendations, and evidence, which is broader than any single REST scope, so a key its creator restricted to specific scopes is refused rather than quietly granted more than it was given. The refusal looks like an ordinary authorization failure with no explanation, so if a key works against the REST API but not here, check its scopes first. Use an unscoped key, or connect over OAuth.

Claude Code (CLI)

Add the server with one command:

claude mcp add --transport http citehawk https://www.citehawk.com/api/mcp \
  --header "Authorization: Bearer YOUR_API_KEY"

Then ask Claude anything about your visibility. For the agentic find-fix-measure loop, see the Claude Code skill.

claude.ai and Claude Desktop

Add a custom connector in your Claude settings pointing at https://www.citehawk.com/api/mcp. Claude will take you through the CiteHawk authorization screen, where you pick the workspace to grant read-only access to, and will list the CiteHawk tools once connected. No header is required for this path.

If you would rather use a key, set the Authorization: Bearer YOUR_API_KEY header instead, using an unscoped key.

Cursor and other MCP clients

Any client that supports a streamable HTTP MCP server works. Point it at the same endpoint with the same Authorization header. In Cursor, add it under Settings > MCP as an HTTP server.

Available tools

Six read-only tools cover the whole picture:

Tool	Returns	Example prompt
`get_visibility_summary`	Latest Visibility and Brand Health scores, sub-scores, KPIs, and which AI engines currently cite you	“How is my AI visibility trending?”
`get_rankings`	Competitor share-of-voice leaderboard for your category	“Who is beating me in AI answers right now?”
`get_prompt_performance`	Your best and worst prompts by mention rate	“Which prompts am I losing on?”
`list_recommendations`	Your current recommendations with priority, impact, and effort	“What should I fix first to get cited more?”
`get_source_gaps`	Cited sources where competitors are listed and you are not	“Which directories cite my competitors but not me?”
`get_evidence`	Real AI answers (provider, prompt, excerpt, cited URL, date) citing a domain	“Show me the actual AI answers citing my site.”

A missing or invalid key returns 401 Unauthorized. Prefer raw HTTP? Use the REST API instead.