Install the GTM standard into your AI client.

In three steps, connect Claude, ChatGPT, Cursor, or any MCP-compatible client to a shared, versioned GTM context.

schema v1.0.0

1 · Choose your client

2 · Install configuration

claude_desktop_config.json

{
  "mcpServers": {
    "gtm-ai": {
      "command": "npx",
      "args": ["-y", "@gtmai/mcp"],
      "env": { "GTM_TOKEN": "gtm_ws_••••" }
    }
  }
}

3 · Verify

Test prompt

Using the GTM standard, reconcile the account data my Apollo MCP just returned against the five canonical objects. Reference Company, Account, Person, Signal, and Engagement fields explicitly.

Expected — first line of the response

"Reconciling against the GTM standard v1.0.0 — mapping the Apollo payload onto Company, Account, Person, Signal, and Engagement below."

Tool surface

Open playground →

Our MCP never proxies or reimplements a vendor's tools. It serves the Standard's definitions and vendor mappings, exposes your approved workspace context, and routes signals to approved plays. When an agent needs live account, signal, or engagement data it calls the vendor's own connected MCP directly.

get_standard()Interactive

Returns the five-object Standard schema (Company, Account, Person, Signal, Engagement), versioned, so the agent can interpret whatever a vendor tool returns.

get_vendor_mapping(vendor)Interactive

The field-mapping table for a connected vendor (Apollo first) so the agent can reconcile that vendor's data against the Standard.

get_workspace_context()Interactive

Read-only, capped access to the workspace's approved context — profile, ICPs, personas, and active plays.

list_templates() / get_template(id) / push_template(id, target_tool)Interactive

Browse and format the prompt/template library for the destination AI tool (Claude, ChatGPT, Gemini, Cursor).

recommend_play(signal)Interactive

Matches a signal to an approved workspace play and returns its trigger, workflow, and governance metadata.

generate_campaign_brief(…)Prototype

Returns a representative campaign brief. Live model inference is described in the AI content-generation specification and is not implemented here.

translate_object(vendor, payload)Prototype

Demonstrates runtime reconciliation of live payloads across two or more connected provider MCPs in an active session.

Supported objects

Troubleshooting

Client can't see the serverReload the client after saving the config. Confirm the command is on your PATH.

Auth errorRotate your workspace token in Settings and paste the new value into the config.

Object missingPin a newer standard version in Settings, or check the object's status in the Standard Explorer.