Connect · MCP

Connect GTM Goat to Claude, ChatGPT, Claude Code and Cursor

GTM Goat is the software; CegTec deploys it with you as forward deployed engineering. Over MCP you steer your workspace from the assistant you already use. Sign-in is OAuth in the browser, with no key in the URL.

Book a meeting →

Server URL

One endpoint for all clients (MCP, Streamable HTTP):

https://app.cegtec.net/api/mcp

How the connection works (OAuth)

The first time you connect, sign-in happens in your browser. You give the client neither a password nor a key.

  1. The client calls the server and is told to sign in.

  2. Your browser opens the GTM Goat sign-in page. Sign in with your account.

  3. Choose the workspace the client may access.

  4. Review which client gets what, and consent.

Access level (scopes)

  • mcp: Full access within your role in the workspace: read and write.
  • mcp:read: Read only. The client cannot create or change anything.

A token belongs to one person in one workspace. When access is renewed, the current role is checked again; a demoted role is carried over, never a higher one.

Step by step per client

Claude (claude.ai)

In settings, under Connectors, add a custom connector.

Enter the server URL:

https://app.cegtec.net/api/mcp

Then choose Connect and finish the sign-in in the browser (choose workspace, consent).

ChatGPT

With a paid ChatGPT plan: in settings under Apps & Connectors, switch on Developer Mode and add a custom connector.

Enter the server URL:

https://app.cegtec.net/api/mcp

Sign-in runs in the browser: choose workspace, consent.

Claude Code

Add the server in your terminal:

claude mcp add --transport http gtm-goat https://app.cegtec.net/api/mcp

Then, inside Claude Code, run /mcp, pick the gtm-goat server and finish the sign-in in the browser.

Cursor

Add it to ~/.cursor/mcp.json (global) or .cursor/mcp.json (project):

{
  "mcpServers": {
    "gtm-goat": {
      "url": "https://app.cegtec.net/api/mcp"
    }
  }
}

On first use Cursor opens the sign-in in the browser: choose workspace, consent.

Example prompts

  • Use workspace_schema_get to show me which tables, sequences and playbooks my workspace has.
  • List my sequences and tell me with get_campaign_stats which one earns the most replies.
  • What does my funnel look like for the last 30 days? Use get_funnel and summarise it in five sentences.
  • Use list_leads to show me the most recent replies and suggest whom I should answer first.
  • Use find_companies to find ten companies that fit my ideal customer profile. Do not enrol anyone.
  • Use preview_sequence_for_lead to show me what the sequence would look like for this lead before I approve anything.
  • Which playbooks do I have, and what does get_setup_status say is still missing to get started?

What is read and what is written

Read

  • Your GTM data in the chosen workspace: tables, companies, leads, replies
  • Sequences, playbooks and metrics (funnel, campaigns, costs)
  • The structure of the workspace (schema)

Write (only with scope mcp)

  • Create and change tables, sequences and drafts
  • Enrol leads in sequences and start workflows or agents
  • Everything inside the chosen workspace and within your role; other workspaces are out of reach

Sending always needs approval

The assistant proposes, a human approves. Messages never go out without approval. Approvals, workspace isolation and cost caps are enforced on the server, regardless of the client and of what a prompt asks for.

Ending access

You end the connection in your application (remove the connection or connector); the server accepts the revocation of the access.

Privacy, terms, support

More clients and tools

More clients, the tool list and technical details:

Does this fit your setup?

30 minutes: we look at how your team can steer GTM Goat from the assistant it already uses.

Book a meeting →