Google Meridian MCP Server (google-meridian-mcp)

A Cloud-Agnostic, First-Principles Model Context Protocol (MCP) server for Google Meridian and Marketing Mix Modeling (MMM).

It empowers data scientists and AI assistants (Cursor, Claude Desktop, Antigravity) to create, audit, calibrate, and optimize mathematically sound, causally valid Marketing Mix Models with zero cloud vendor lock-in.

💡 Why First Principles Over Rigid Scripts?

Most assistant implementations rely on hardcoded procedural scripts (SKILL.md). In real-world data science, rigid scripts break because every business, industry, and marketing dataset is unique:

• A retail brand with 50 DMAs operates differently than a B2B SaaS startup with national data.

• App install targets (NON_REVENUE) require different priors than revenue models (REVENUE).

• Specific channel CPMs, flighting patterns, and Lift Tests vary across campaigns.

First Principles never change. Causal identification (DAGs), carryover/saturation physics (Adstock & Hill curves), Bayesian probability calibration, NUTS MCMC geometry ($\hat{R} < 1.05$), and KKT convex budget optimization apply universally to every dataset on Earth.

By anchoring this MCP server in First Principles & Dynamic Data Science Tools, the co-pilot adapts seamlessly to any specific edge case while guaranteeing mathematical rigor.

Related MCP server: Gemini Docs MCP Server

🏛️ The 4 Mandate Pillars

┌────────────────────────────────────────────────────────────────────────────────────────┐
│ 1. EVERY CONTROL POINT   │ Covers Data Inputs, Controls/DAG, Baseline, Adstock,        │
│                          │ Hill Saturation, Bayesian Priors, MCMC, Optimization.       │
├──────────────────────────┼─────────────────────────────────────────────────────────────┤
│ 2. END-TO-END WORKFLOW   │ Guides the 5 phases & 3 iteration loops (Convergence        │
│                          │ Diagnostics ➔ Causal Plausibility ➔ Lift Calibration).      │
├──────────────────────────┼─────────────────────────────────────────────────────────────┤
│ 3. FIRST PRINCIPLES      │ Grounded in causal inference, probability theory, HMC/NUTS  │
│                          │ sampling geometry, and convex optimization math.            │
├──────────────────────────┼─────────────────────────────────────────────────────────────┤
│ 4. PLATFORM AGNOSTIC     │ 100% portable with zero cloud vendor lock-in. Runs locally, │
│                          │ in Docker, or on AWS, Azure, GCP, Railway, Render, etc.     │
└────────────────────────────────────────────────────────────────────────────────────────┘

🛠️ Complete MCP Tool Suite (11 Tools)

⚡ Quick Connect (Remote SSE Mode)

Add this to your IDE's mcp_config.json:

{
  "mcpServers": {
    "google-meridian": {
      "url": "https://google-meridian.mcp.borobudur.ai/sse"
    }
  }
}

💻 Local Setup & Running

Option A: Python FastMCP Setup (Recommended)

# Install dependencies
pip install -r requirements.txt

# Run server in stdio mode
python server.py

mcp_config.json:

{
  "mcpServers": {
    "google-meridian-mcp": {
      "command": "python",
      "args": [
        "C:/path/to/google-meridian-mcp/server.py"
      ]
    }
  }
}

Option B: Node.js Setup

npm install
npm start

📄 License

MIT License. Open source and free for the community.