Skip to main content
The last Setup step lets you connect Cometly to your AI agent, so your Cometly data sits behind the AI tools you already use. It has two tabs, one for each way of doing that: MCP gives an assistant or coding agent live tools to query your space (a space is a workspace inside your Cometly account, with its own settings, integrations, data and members), and Data Warehouse sends the data itself to storage you own. MCP (Model Context Protocol) is an open standard that lets AI tools such as Claude, ChatGPT and Cursor call tools on another service. Once a tool is connected, you can ask questions about your ad performance in plain language, for example “What was my ROAS by source last month?”, and the assistant answers from your space’s live data. A warehouse export suits teams that want the raw data in their own storage to combine with other sources.

Where to find it

In the left sidebar, click Setup, then open step 7, Add Cometly to your AI agent.

MCP tab

  • Your MCP server URL, with a Copy button. There is no API key: you add the URL to your tool, sign in to Cometly from it, pick the space and approve the access.
  • Guides: choose your tool (Claude, ChatGPT, Grok, Henry, Cursor, Claude Code, Gemini CLI, VS Code or Windsurf) to see its steps and, under What to copy, the Server URL plus a command or config snippet for the tools that use one. Selecting ChatGPT swaps the URL for the one ChatGPT needs.
  • What you can do: example prompts for an AI assistant and for a coding agent.
  • Server-side agents and your own integration: create an access token for an agent that cannot sign in through a browser. See Authentication.
The Connect Claude guide on the MCP tab, with its three steps on the left and the What to copy card showing the Server URL and a copy button on the right.
Anyone who belongs to the space can connect. A connection made by a Viewer is read-only.

Data Warehouse tab

  • Set up your first export (Manage exports once you have one), next to a count of your destinations and active exports.
  • Datasets you can export.
  • Destinations: Amazon S3, Google Cloud Storage, Snowflake and BigQuery, each with a Set up or Manage button.
Setup step 7 on the Data Warehouse tab in a space with no exports, showing the Set up your first export button, the datasets you can export and the Destinations card with a Set up button for each destination.
Any member of the space can view destinations and exports. Only a space Administrator can add or change them.

Check it worked

You cannot tick this step by hand. Its badge shows Connected when either of these is true:
  • An AI tool connected over MCP has asked Cometly something about this space in the last 30 days.
  • At least one data export is active.
Approving access is not enough on its own. Until your agent makes its first request the badge reads Authorised, no calls yet, and before that Not connected. This is the last step, so its button is Finish setup, which opens your dashboard.

Frequently asked questions

Can an AI agent change my Cometly data?

Mostly no. Nearly every tool the agent gets is read-only. Three tools can edit fields on a single contact, company or event, and no tool can create or delete data or change attribution. A connection made by a Viewer has no write tools at all.

Can I connect more than one Cometly space to the same AI tool?

Yes. When you approve the connection, you choose one space, specific spaces or All my spaces, plus a default space. The agent works in the default space unless it names another space the connection covers.

How do I disconnect an AI tool from Cometly?

Open Settings → Integrations → MCP, find the tool’s token under Server-side (headless) agents and click Revoke. Connections you approved from a tool such as Claude or Cursor appear in that list too, named after the tool. The tool’s next request is refused until you connect it again.

Next steps

Still have questions or need a hand? We’re happy to help. Reach our team at support@cometly.com.