Marketing Analytics MCP Server

A Model Context Protocol server that gives an LLM live, read-only access to a marketing analytics stack — so you can ask questions in natural language instead of jumping between dashboards. Exposes tools across Google Search Console, GA4, Google Ads, HubSpot CRM, and Bing Webmaster Tools, plus composite cross-platform rollups.

Runs locally over stdio (Claude Desktop / Claude Code) or as a remote streamable-HTTP service (e.g. Cloud Run) with bearer-token auth.

Tools

Area	Examples
Search Console	queries, pages, CTR/position, period comparison, movers & losers
GA4	sessions, channels, landing pages by source, funnels, conversions
Google Ads	campaign performance, keyword quality, search terms
HubSpot	deal / company / contact search, pipeline summary, activity timelines
Bing	top queries/pages, branded vs unbranded, crawl stats
Composite	cross-platform marketing snapshot

Design

• clients/ — one thin async client per platform (auth + raw calls).
• tools/ — MCP tool definitions that shape client output for an LLM.
• cache.py — TTL cache to stay inside free-tier API limits.
• circuit_breaker.py — trips on repeated upstream failures so one bad provider can't hang the server.
• server.py — registers tools and serves stdio or streamable-HTTP.

Setup

pip install -e .
cp .env.example .env     # fill in your own credentials + property IDs
python -m mcp_server.server

All credentials and property/account IDs come from environment variables (see .env.example) — nothing is hardcoded. You point it at your own analytics accounts.

Deploy (remote)

A Dockerfile + Procfile are included. Set MCP_AUTH_TOKEN so only requests carrying the bearer token can connect, then deploy to any container host.

Tests

pytest