td_live_… token that appears. This is the only time it's shown. Store it in a password manager.read:metrics can only read metrics. Even if the LLM tries to call run_probe, the tool won't be visible. Pick the least-privilege set that covers your workflow.After issuing, the dashboard shows a Connect an agent panel with one-click install snippets per client. You can also paste the snippets below directly.
// ~/Library/Application Support/Claude/claude_desktop_config.json (macOS)
// %APPDATA%\Claude\claude_desktop_config.json (Windows)
{
"mcpServers": {
"trustdata": {
"url": "https://app.trustdata.tech/mcp/",
"headers": {
"Authorization": "Bearer td_live_YOUR_TOKEN"
}
}
}
}
claude mcp add --transport http trustdata https://app.trustdata.tech/mcp/ \
--header "Authorization: Bearer td_live_YOUR_TOKEN"
// ~/.cursor/mcp.json
{
"mcpServers": {
"trustdata": {
"url": "https://app.trustdata.tech/mcp/",
"headers": { "Authorization": "Bearer td_live_YOUR_TOKEN" }
}
}
}
// ~/.gemini/settings.json
{
"mcpServers": {
"trustdata": {
"httpUrl": "https://app.trustdata.tech/mcp/",
"headers": {
"Authorization": "Bearer td_live_YOUR_TOKEN"
}
}
}
}
Settings → Connectors → Add MCP server
URL: https://app.trustdata.tech/mcp/
Auth: Bearer token
Token: td_live_YOUR_TOKEN
Restart the agent. You should see a trustdata connection in its MCP / tools panel.
Ask your agent something that exercises a tool:
"List the TrustData analytics queries available to me."
The agent will call list_analytics_queries and return the set of query types your token's scopes allow. If you see a list of query types (campaigns, channel_summary, geo_visibility, etc.), you're connected.
The real value is letting the agent drive an end-to-end loop. Open the SEO × GEO playbook and paste it as a system prompt. The agent will diagnose, propose hypotheses, and record trackable recommendations in one conversation.
| Tier | Rate cap | Best for |
|---|---|---|
| Measure | 30 requests/min | Manual exploration, one-off questions |
| Optimize | 300 requests/min | Recurring reports, dashboards |
| Prove | 3000 requests/min | Agentic workflows (recommended) |
run_probe is separately metered against your GEO add-on's monthly engine-call allowance.
| Symptom | Cause | Fix |
|---|---|---|
403 SCOPE_DENIED | Token is missing the scope the tool requires | Issue a new token with the needed scope (tokens are immutable, so issue a replacement and revoke the old one). |
404 on property_id | Property is in a different org, or was deleted | Confirm the property UUID in the dashboard URL. |
422 GUARDRAIL_VIOLATION | Hypothesis shorter than 20 chars, or invalid target_prompt_ids | Write a proper "If we X, then Y because Z" hypothesis. Verify prompt IDs belong to your org. |
| Agent doesn't see TrustData tools | Token lacks matching scopes, so filtered tools are hidden entirely | Add scopes, reconnect the MCP server. |
401 UNAUTHENTICATED after it worked before | Token was revoked | Issue a new one in Developer → API tokens. |