2.3 · Connect your agent via MCP
Connect your AI tools to your workspace so an agent can pull scores, traffic and briefs directly.
MCP is how your AI tools get direct access to your GrowthOS workspace, the container holding everything GrowthOS knows and does for your brand. Once connected, your agent can pull any page's scores and traffic, read a brief (the plan for one page) and its research, check your context (what GrowthOS knows about your company), and answer portfolio questions, all live from the workspace rather than from whatever you paste into the chat. It's the foundation for publishing via your agent and for the playbook of tasks later in the tutorial.
The connection is yours personally: you sign in with your own GrowthOS account, and your agent gets read-only access to the workspaces you can already see. Out of the box, nothing your agent does over MCP can change your workspace; letting it make changes is a separate, deliberate grant, covered in 8.2 · Give your agent write access.
Where setup lives
In the top bar, open the ⋯ menu and choose MCP under Lab. The MCP Settings page carries the connection URL, per-tool instructions, and a Connected badge once your first request lands. Pick your tool and the steps adjust; here's the short version of each.
Claude (web)
Add the connector
In Claude, open Customize › Connectors › Add custom connector, and paste the connection URL from your MCP Settings page.
On a Claude Team or Enterprise plan, connectors are admin-managed: you won't be able to add a custom connector yourself. Ask a Claude workspace Owner to add it under Organization settings › Connectors, using the connection URL from your MCP Settings page. The Owner also chooses the connector's display name, so it may not be called "GrowthOS" in your list; identify it by the connection URL. Once it's added, the next step is the same for everyone: each person connects and signs in individually.
Connect and sign in
Select Connect on the GrowthOS connector and sign in with your GrowthOS account. Approve the access prompt and you're done.
Claude Code
Register the server
Run the claude mcp add command shown on your MCP Settings page (it registers the growthx-os server with your connection URL).
Authenticate
In a new session, run /mcp, select growthx-os, choose Authenticate, and sign in with your GrowthOS account.
Cursor
Add the server
Open Settings › Tools & MCPs › New MCP server and paste the mcp.json snippet from your MCP Settings page.
Authorize
Click Connect next to growthx-os and sign in.
Any other MCP-capable tool works the same way: add a server with the connection URL, then sign in with your GrowthOS account when prompted.
Confirm it works
Ask your agent something only your workspace can answer: "Using GrowthOS, what are my top pages by traffic this month?" A real answer means you're connected, and the MCP Settings page shows its Connected badge. The same page lists everything your agent can do, under What your agent can do: finding and auditing pages, traffic and keyword data, briefs and opportunities (proposed pages worth writing or rewriting), and your context documents, all read-only by default.
Common questions
Does each teammate connect separately? Yes. Credentials are per person, so each teammate signs in with their own account and gets access to the workspaces they already have.
Can my agent change or delete anything through this? Not unless you deliberately allow it. The default connection is read-only; agent-drafted publishing happens through your CMS's own integration with the article read from GrowthOS (5.4 · Polish and publish covers it). Granting your agent the ability to change the workspace itself is an explicit opt-in on the sign-in screen, covered in 8.2 · Give your agent write access.
Is this safe to roll out to the whole team? It grants each person's agent the same visibility that person already has, read-only unless they individually opt into write access. If someone leaves, removing their GrowthOS seat ends their access, MCP included.