cli

@marketingmcp/cli

Local stdio MCP server for MCP Ads. Use it with Claude Code, Cursor, Cline, Continue, Windsurf, or any MCP-compatible agent — no Python needed on the client machine.

It works against either backend:

Cloud (mcp-ads.com)

1. Get an API key at https://mcp-ads.com/manage (CLI API Keys section).
2. Connect your accounts (Google Ads, GA4, GSC, Meta Ads, …) on the same dashboard — no platform API applications needed on your side.
3. Run:

MCP_ADS_API_KEY=mcp_live_YOUR_KEY npx -y @marketingmcp/cli

Self-hosted (this repo's server)

Run the open-source server in HTTP mode (python server.py serve --transport http), then point the CLI at it:

MCP_ADS_BASE_URL=http://127.0.0.1:8000 npx -y @marketingmcp/cli

No API key needed on loopback. If you set MCP_ADS_API_KEY on the server, pass the same value to the CLI.

How it works

The CLI exposes ~17 category meta-tools (one per platform area) plus a list_actions discovery tool. Each tool call is forwarded over HTTP to the backend's /api/cli/dispatch, which resolves the right credentials and runs the action.

This means:

• Your AI agent loads ~18 tool definitions instead of 1,000+ — dramatically less context spent on tool listing.
• Credentials stay on the server — the CLI never sees them.
• Cloud keys revoked at https://mcp-ads.com/manage stop working immediately.

Config

Source	Variable	Default
Env	MCP_ADS_API_KEY	— (required for cloud; optional self-hosted)
Env	MCP_ADS_BASE_URL	https://mcp-ads.com
CLI flag	--api-key, -k	(overrides env)
CLI flag	--base-url	(overrides env)

Development

npm install
npm run build       # produces bin/marketingmcp.js
npm run typecheck

To smoke-test against a local backend:

MCP_ADS_BASE_URL=http://localhost:8000 node bin/marketingmcp.js

Or drive it from the MCP Inspector:

npx @modelcontextprotocol/inspector node bin/marketingmcp.js