Skip to main content
Trodo’s MCP (Model Context Protocol) server lets AI clients query your event analytics, agent runs, evals, issues, and clusters in natural language — without leaving the chat or IDE.

​
What you get

A single server at https://mcp.trodo.ai/mcp exposing 70+ tools across 8 capability buckets:
  • Analytics — funnels, retention, insights / DAU-MAU trends, segment comparisons, UTM attribution, cohort comparisons, paths, flows, anomalies, period-over-period diffs.
  • Agent runs — list / search / drill into LLM-agent runs and span trees; tool-call analysis, token cost, top failure modes, feedback summaries.
  • Evals — list / inspect evaluators; fetch eval results for a run or span; create / update / toggle / delete / test / backfill evaluators; submit, skip, or list pending human grades.
  • Issues — detected tool-failure / output-quality / UX-rage / conversation-breakdown issues; details, timeline, members, top-failing-tools and rage-hotspot shortcuts; status changes.
  • Identity (PII, opt-in) — user profile, journey, search by email / name / wallet; users-who-fired-event, users-affected-by-issue.
  • Groups — list and inspect customer groups / accounts / workspaces.
  • UX & health — rage clicks, scroll depth, exit intent, form abandonment, JS / network errors, page performance, error-impact-on-funnel.
  • Reports & catalog — export saved reports; list event names / properties / values, agent names, run dimensions.
Each tool’s description carries a [BUCKET] prefix and meta.x-trodo-bucket annotation, and the server’s top-level instructions block ships a one-paragraph routing guide — so the host LLM picks tools accurately without trial-and-error. Identity-style params (emails, names, free-text run / cluster / issue references) are resolved server-side: pass natural language, get the right ID under the hood. No need to look up UUIDs. All scoped to the team you connect with. No data leaves your team.

​
Pick your client

You’re usingConnect viaSee
Claude.ai (web)OAuthClaude.ai quickstart
Claude DesktopOAuthClaude Desktop quickstart
Claude Code (CLI)API key (HTTP or stdio)Claude Code quickstart
CursorAPI keyCursor quickstart
Custom Python / Node scriptAPI keyCustom client quickstart

​
Plan requirement

MCP is included on Growth and Enterprise plans. Free teams see an upgrade prompt on the consent screen and cannot create API keys.

​
What you should know

  • Auth: OAuth 2.1 with PKCE for browser clients; per-team API keys (trodo_mk_*) for headless clients.
  • Scopes: every tool requires a scope. The PII scope (mcp:user:read_pii) is off by default — opt in only if you want Claude to read identity data. See Scopes.
  • Rate limits: per-token-per-minute caps; per-tool concurrency caps for heavy queries. See Rate limits.
  • Privacy: see Data handling for what’s stored and what’s sent to the AI client.

​
Manage MCP for your team

In the dashboard: Integrations → MCP for Claude & Cursor. Create / revoke API keys, view connected OAuth grants, copy per-client config snippets.