Jarvis: marketing MCP server for Claude

One self-hosted MCP server that gives Claude read and write access to your marketing stack: about 216 tools across Google Ads, Meta Ads, GA4, Google Search Console and Microsoft Clarity.

Use it locally from Claude Code or Claude Desktop, or host it and add it to claude.ai as a custom connector, so it also works in the Claude mobile app and in Cowork.

What's inside

Safety: every change is preview-first

Every tool that changes an account takes confirm=false by default. That call changes nothing: it asks the platform to validate the change (Google Ads validate_only, Meta execution_options=validate_only) and returns a preview. Claude is instructed to show you the preview and only call again with confirm=true after you approve. New Meta campaigns, ad sets and ads are created paused.

Related MCP server: meta-ads-mcp-server

Setup

Requires Python 3.11+.

git clone https://github.com/cgerig/jarvis-marketing-mcp.git
cd jarvis-marketing-mcp
python -m venv .venv && .venv/bin/pip install -r requirements.txt
cp .env.example .env   # then fill it in

Credentials

You only need the platforms you use; tools for a platform without credentials return a readable error.

• Google (Ads, GA4, Search Console, Customer Match): sign in with Application Default Credentials:

  gcloud auth application-default login --scopes=https://www.googleapis.com/auth/adwords,https://www.googleapis.com/auth/analytics.readonly,https://www.googleapis.com/auth/webmasters,https://www.googleapis.com/auth/datamanager,https://www.googleapis.com/auth/cloud-platform

  Google Ads also needs a developer token. Enable the Google Ads, Analytics Data, Analytics Admin, Search Console and Data Manager APIs on your Cloud project.

• Meta: a System User access token (Business Settings > Users > System users) with ads_management, ads_read, business_management, leads_retrieval, pages_show_list, pages_read_engagement, pages_manage_ads and read_insights.

• Clarity: Settings > Data Export > Generate new API token (Clarity allows 10 export calls a day).

Use it

Locally (Claude Code or Claude Desktop)

Runs over stdio, so nothing is exposed to the network:

claude mcp add jarvis --scope user -- /path/to/jarvis-marketing-mcp/run_stdio.sh

As a remote connector (claude.ai, mobile, Cowork)

connector/app.py serves everything over Streamable HTTP at /mcp, protected by GitHub sign-in that only lets in the usernames in ALLOWED_GITHUB_USERS.

1. Pick a public HTTPS address: a host such as Cloud Run, Render or Fly.io (use the Dockerfile), or your own computer behind Tailscale Funnel or Cloudflare Tunnel.

2. Create a GitHub OAuth App with callback URL https://YOUR-ADDRESS/auth/callback.

3. Set the PUBLIC_URL, GITHUB_*, ALLOWED_GITHUB_USERS and JWT_SIGNING_KEY settings (see .env.example), and point OAUTH_STORE_DIR at persistent storage so sign-ins survive restarts.

4. Start it with ./run_http.sh, or deploy the Docker image.

5. In claude.ai: Settings > Connectors > Add custom connector, URL https://YOUR-ADDRESS/mcp.

Keep secrets in .env or your host's secret manager; .gitignore already excludes them.

Project layout

connector/app.py      combined server: mounts the three tool sets, sign-in, stdio or http
google/gads_tools/    Google Ads tools
google/gx_tools/      GA4 extras, Search Console, Clarity
meta/meta_tools/      Meta Ads tools

License

MIT