Connect Claude, Cursor, or ChatGPT via MCP

You want to ask your AI Peekaboo data questions in plain language, or automate bulk work like creating dozens of prompts at once, directly from Claude or Cursor, instead of clicking through the dashboard. (ChatGPT and Codex connect through the API directly; see "Other AI tools" below.) MCP is the bridge that lets those tools reach your data over the same API key you would use for the REST API, so the answers come from your real account.

What this gets you

AI Peekaboo runs an MCP server that lets Claude Desktop, Claude Code, Cursor, or any MCP-compatible AI client connect straight to your own workspace data using your API key. Once connected, you can ask it things like "what's our visibility score this month" or "which competitors are outranking us on our commercial prompts" and get real answers pulled live from your account, not guesses.

The server exposes 19 tools: 16 read operations and 3 write operations. A read-only key can use the read operations, subject to your plan and local-tracking rollout, so you can:

• Pull brand and visibility data
• Read competitor data
• List cited sources and URLs
• Read prompts and prompt detail
• Check your API rate-limit status
• Get recommended prompts
• Read local settings and reports, search city targets and preview model coverage

With a read+write key, you can also:

• Add new prompts
• Add new competitors
• Enable, change or pause eligible local tracking

Setting it up

1. Generate an API key from Settings → Integrations in AI Peekaboo. Use a read+write key if you want to create prompts or competitors, or configure local tracking through the connection, or a read-only key if you just want to query data.
2. Add AI Peekaboo as an MCP server in your client of choice (Claude Desktop, Claude Code, or Cursor), pointing it at AI Peekaboo's MCP endpoint and passing your API key as an environment variable.
3. Restart or reload the client so it picks up the new server.
4. Ask it a question about your brand to confirm the connection works.

Setting environment variables for MCP tends to need more care on Windows than on Mac.

Rate limits

Every plan includes a per-minute and a daily allowance for API and MCP requests, and every request your AI tool makes counts toward it, including the connection handshake, not only the questions you ask. When you reach the limit, tool calls return a "rate limit reached" message that says when it resets (per-minute limits reset within a minute; daily limits reset at midnight UTC). Your connection is still valid, so there's no need to sign in again, just wait for the reset or upgrade your plan for a larger allowance. You can ask your AI tool to run the get_rate_limit_status tool at any time to see how much of your allowance is left.

Practical ways people use this

• Bulk-creating prompts and competitors. The dashboard UI is built for adding one at a time. If you're filling out a new brand with dozens of prompts and a full competitor set, connecting through MCP with a read+write key and asking Claude or Cursor to create them in bulk is dramatically faster than doing it by hand. Brands themselves aren't creatable over MCP, so set the brand up in the dashboard first, then point the connection at it.
• Building custom client reports. Agencies use the same connection to pull live data into a custom report format, whether that's a one-off summary or a repeatable template you reuse per client.

Other AI tools

ChatGPT and Codex can also connect to AI Peekaboo's API directly and work the same way, using your API key.

Gemini does not have a native app integration today. If you're working in Gemini, use CSV export from the dashboard as your workaround instead.

Local tracking with your AI agent

Use list_local_tracking to find a brand's configurations and get_local_tracking for a prompt's settings, version and historical target/model pairs. Ask get_local_report for one explicit target, model and date window. For example: "Show this prompt's local results for the saved target and Google AI Mode over 14 days. Include the recorded targeting precision and any missing evidence. Keep this separate from my standard visibility score."

For a change, the agent must read settings, use search_local_targets, then call get_local_coverage for the chosen target. Review each model's city, state, country or unavailable coverage before it calls set_local_tracking with the current version, coverage fingerprint, a fresh UUID idempotency key and any required acknowledgement. Stopping uses the same write tool with targetId: null and requires a version and idempotency key, but no coverage preview.

A read+write key is required for changes. If a response says coverage or configuration changed, reread it before confirming another change. Adding a location to add_prompt is rejected; create the prompt first, then configure its local tracking. See local tracking limits for report bounds and support details.