Model Context Protocol (MCP) Integration

Ask Claude, ChatGPT, Cursor or another MCP client about your SourceTrack setup and your counts without leaving the chat. SourceTrack’s Model Context Protocol (MCP) server answers with numbers SourceTrack has already computed for your site.

Why use SourceTrack MCP with AI assistants

Plain-language questions: Ask “Is the tracker firing?”, “Which campaigns brought the most leads last month?” or “Which of my conversion destinations failed its last send?”
Real numbers only: Every tool returns values SourceTrack already computed for your site. Where a value cannot be measured, the tools leave it out or return null rather than show 0.
One token, one site: The site is resolved from your API token. No tool takes a site argument, so an assistant cannot be talked into reading another site.
No writes: Every documented tool reads. The three groundwork write tools run in dry-run mode only and change nothing.
Personal data stays masked: Contact ids are masked, deal names are masked as names, and email addresses are masked.

What tools are available

The full list, with the exact description your assistant is given, is on the MCP documentation page.

Setup and health: detect_platform, get_install_snippet, verify_installation, get_workspace_context, get_site_health, get_data_quality, debug_data_flow, verify_events, explain_metrics.
Conversion destinations: list_active_destinations shows which Conversion API destinations (Meta, Google Ads, GA4, TikTok and others) are set up and how their newest sends went.
Leads, campaigns and contacts: get_leads_volume, get_campaign_volume, find_contacts, get_journey.
CRM deals and pipeline: list_deals, get_pipeline_stages.
Conversions by channel: get_conversions_by_channel (first-touch counts) and get_paths (conversion path counts).
Traffic and funnels: get_traffic_overview, get_traffic_timeseries, get_traffic_breakdown, list_saved_funnels, get_funnel_results.
Groundwork (dry run only): push_offline_conversion, sync_lead_to_crm, tag_campaign.

The documented tools report counts and setup state; revenue, cost, ROAS, CPL and CAC reports are not among them. No tool lets the assistant choose an attribution model: where a tool uses a touch, it is always the first touch.

Step-by-step setup in under 2 minutes

Step 1: Create an API token

In the SourceTrack dashboard, open Settings, then Advanced, then API tokens, and generate a token with only the scopes you need.

Step 2: Add the MCP server

Add this block to claude_desktop_config.json, ~/.cursor/mcp.json or your client’s MCP settings:

{
  "mcpServers": {
    "sourcetrack": {
      "command": "npx",
      "args": ["-y", "@sourcetrack/mcp"],
      "env": {
        "SOURCETRACK_API_KEY": "st_live_your_api_token_here"
      }
    }
  }
}

Remote clients such as ChatGPT connect to the hosted endpoint https://api.srctk.com/api/mcp with the token in an Authorization: Bearer header. See the MCP documentation for every client.

Step 3: Start asking

Open your assistant and ask: “Using SourceTrack, how many leads did each campaign bring in the last 30 days?”

Frequently Asked Questions

Can AI models modify or delete my data?

No. The documented tools read only. The three groundwork tools return a dry-run preview and refuse a live write.

Is sensitive customer information shared with the AI?

Contact ids are masked, deal names are masked as names, and email addresses are masked. No tool returns a visitor’s IP address.

Model Context Protocol (MCP) integration