Your AI visibility data,
in your own tools.

An owner or admin creates an API key in Settings → API & agents. It is shown once; Optimz stores only its hash. Send it as Authorization: Bearer optimz_live_…. AI apps like Claude and ChatGPT connect with OAuth instead: a person signs in and picks the agency and the access level.

Every call runs with the same checks as the app: your role, plan limits, daily limits and the plan's monthly usage limit. A key acts as the admin who created it and stops working when it is revoked or that person stops being an admin; an app acts as the person who connected it, within their current role. Every change made through the API is listed in the agent activity log in Settings. REST calls are server to server: never put a key in a web page.

Base URL https://www.optimz.com/api/v1. JSON only. The full schema is the OpenAPI document.

curl -H "Authorization: Bearer $OPTIMZ_API_KEY" \
  https://www.optimz.com/api/v1/clients

curl -H "Authorization: Bearer $OPTIMZ_API_KEY" \
  "https://www.optimz.com/api/v1/clients/<client-id>/visibility?weeks=4&engine=chatgpt"

Lists return { data, next_cursor }: pass cursor=<next_cursor> with the same filters for the next page (limit 1–100, default 50). Actions list the top 500 and content the newest 200.

Add a question (write access):

curl -X POST -H "Authorization: Bearer $OPTIMZ_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"text": "What is the best CRM for a 10-person sales team?"}' \
  https://www.optimz.com/api/v1/clients/<client-id>/questions

Creating an article (POST …/content with a question_id or action_id) counts against the client's monthly article cap each time it succeeds, so don't blindly retry a 201. A person still reviews and approves every article in Optimz before anything is published.

Every rate carries what it was computed from: k of n answers, the proportion, and its Wilson 95% confidence interval. With no answers, value is null, not 0%.

"mention": {
  "k": 41,
  "n": 72,
  "value": 0.569,
  "ci_low": 0.455,
  "ci_high": 0.677,
  "ci_label": "95% CI 46–68%",
  "reliability": "early_estimate",
  "plain": "Based on 72 answers · Early estimate"
}

Engines are named with what was measured, e.g. ChatGPT · API with web search, Google AI Overviews · Google results page, or Simulated for demo data. Answers where the surface showed no AI answer are listed with surface_shown: false and never counted.

Ask your agent "how is our AEO?", "what should we fix first for Acme?" or "add printthesheet.com as a client". The server is https://www.optimz.com/api/mcp (streamable HTTP).

Claude: Settings → Connectors → Add custom connector, paste the URL and sign in. ChatGPT: Settings → Connectors → Create, paste the URL, OAuth. Sign-in uses OAuth 2.1 with PKCE and dynamic client registration (discovery at /.well-known/oauth-protected-resource/api/mcp).

Claude Code, with an API key:

claude mcp add --transport http optimz https://www.optimz.com/api/mcp \
  --header "Authorization: Bearer $OPTIMZ_API_KEY"

Claude Desktop (claude_desktop_config.json):

{
  "mcpServers": {
    "optimz": {
      "command": "npx",
      "args": [
        "-y", "mcp-remote", "https://www.optimz.com/api/mcp",
        "--header", "Authorization: Bearer ${OPTIMZ_API_KEY}"
      ],
      "env": { "OPTIMZ_API_KEY": "optimz_live_…" }
    }
  }
}

Cursor (.cursor/mcp.json):

{
  "mcpServers": {
    "optimz": {
      "url": "https://www.optimz.com/api/mcp",
      "headers": { "Authorization": "Bearer optimz_live_…" }
    }
  }
}

Destructive tools (archiving, revoking a share link) first answer with a confirmation token and what they would do; the agent calls again with confirm once you agree. Data sources that need a sign-in (Google, Microsoft, your host) are connected by a person: check_connections returns the link.

✱ asks for confirmation first.

Errors are { "error": { "code", "message" } }. Limits per key or connected app: 120 requests and 10 writes per minute; bodies up to 64 KB.