Engageable Analytics

The open source analytics engine for AI agents.

One MCP server that connects to GA4, Mixpanel, PostHog, and more. 9 tools that replace per-platform integrations. Bring your own Anthropic key.

Quickstart

pip install engageable
export ANTHROPIC_API_KEY=sk-ant-...
export POSTHOG_API_KEY=phc_...
export POSTHOG_PROJECT_ID=12345
engageable-mcp

That's it. The MCP server is running on stdio. Connect it to Claude Desktop, Cursor, or any MCP client.

Claude Desktop

Add to ~/Library/Application Support/Claude/claude_desktop_config.json:

{
  "mcpServers": {
    "engageable": {
      "command": "engageable-mcp",
      "args": []
    }
  }
}

Restart Claude Desktop. You'll see 9 analytics tools available.

Docker

ANTHROPIC_API_KEY=sk-ant-... docker compose -f docker-compose.mcp.yml up

Connects via SSE at http://localhost:8080/sse.

Tools

Tool	What it does
get_sources	List connected data sources
configure_source	Connect a new data source (saves to ~/.engageable/credentials.json)
analyze_trends	Time-series analysis with trend detection, change points, anomalies
compare_segments	A/B tests, before/after, segment breakdown with statistical significance
detect_anomalies	Find spikes, drops, and unusual patterns
analyze_retention	Cohort retention curves (D1/D7/D30)
analyze_funnel	Multi-step conversion funnel with drop-off rates
analyze_cohort	Define and compare user cohorts
ask	Natural language analytics questions (routes to other tools via LLM)

Supported Data Sources

Source	Auth	What you need
PostHog	API key	POSTHOG_API_KEY + POSTHOG_PROJECT_ID
Mixpanel	Service account	MIXPANEL_SERVICE_ACCOUNT_USERNAME + MIXPANEL_SERVICE_ACCOUNT_SECRET + MIXPANEL_PROJECT_ID
Google Analytics 4	Service account	GA4_CREDENTIALS_JSON (path or inline) + GA4_PROPERTY_ID

Set these as environment variables, or use the configure_source tool to save them interactively to ~/.engageable/credentials.json. See credentials.example.json for the file format.

How It Works

Engageable exposes analytics tools via the Model Context Protocol (MCP). Any MCP-compatible client (Claude, Cursor, VS Code, custom agents) can discover and call these tools.

Each tool is a self-contained pipeline: parse the request, fetch data from the right connector, run analysis, return results. The ask tool adds an LLM routing layer for natural language questions.

Responses use CSV for tabular data (50% fewer tokens than JSON) with a 1000-cell budget to keep context windows manageable.

Architecture

MCP Client (Claude, Cursor, etc.)
    │
    ▼
┌─────────────────────────────────────┐
│  MCP Server (stdio or SSE)          │
│  - Dynamic tool registration        │
│  - Credential injection             │
│  - CSV response formatting          │
├─────────────────────────────────────┤
│  Composite Skills (agent-facing)    │
│  analyze_trends, compare_segments,  │
│  detect_anomalies, analyze_funnel,  │
│  analyze_retention, analyze_cohort, │
│  ask, get_sources, configure_source │
├─────────────────────────────────────┤
│  Connector Skills (internal)        │  Analysis Skills (internal)
│  ga4_query, posthog_query,          │  trend_detection, significance,
│  mixpanel_query, + metadata/probe   │  cohort_retention, forecasting,
│                                     │  bayesian_ab, correlation, ...
└─────────────────────────────────────┘

License

MIT