概览
What is Google Ads Intent MCP?
Google Ads Intent MCP is a dry-run-first search-term intent analyzer for agents that classifies Google Ads search terms and drafts negative-keyword plans without making live account changes. It is built for MCP clients like Codex, Claude, Cursor, Hermes, and OpenClaw, helping reduce wasted spend while protecting buyer-intent queries.
How to use Google Ads Intent MCP?
Install via pipx: pipx install google-ads-intent-mcp (or pipx install "google-ads-intent-mcp[mcp]" for MCP support). Then run the MCP server with google-ads-intent-mcp. Recommended first calls: google_ads_connection_status, google_ads_privacy_audit, google_ads_analyze_search_terms, and google_ads_build_negative_plan. The CLI offers commands like google-ads-intent manifest, doctor, privacy-audit, classify, analyze-csv, and plan-negatives.
Key features of Google Ads Intent MCP
- Dry-run analysis of exported search-term CSVs
- Classifies intent into waste, buyer, research, or competitor
- Drafts negative-keyword plans without applying them
- Privacy audit and connection status introspection tools
- Deterministic heuristic with optional LLM refinement
- Buyer/conversion terms automatically protected from exclusion
Use cases of Google Ads Intent MCP
- Reduce wasted ad spend by identifying non-converting search terms
- Safely generate negative keyword plans before making live changes
- Protect buyer-intent queries from accidental exclusion
- Automate search-term analysis in agent-driven ad workflows
- Local CSV analysis without exposing account credentials to agents
FAQ from Google Ads Intent MCP
What does Google Ads Intent MCP do differently from alternatives?
It is dry-run-first: it never mutates Google Ads accounts directly. CSV analysis is local, and negative plans are drafted but not applied, making the safe path the default.
What are the runtime requirements?
Python 3.12+ with pipx. No API keys are required for the default heuristic classifier. An optional LLM refinement path requires OPENAI_API_KEY and the openai package.
Where does the data live?
All CSV analysis is performed locally. Search terms, costs, and clicks are processed on the user’s machine. OAuth tokens and developer tokens should remain in the local environment.
What are the known limits?
The current version (v0.1) does not perform live mutations. It only analyzes CSV exports. Negative keyword plans are dry-run only.
What transport and authentication does it use?
The MCP server uses stdio transport. It does not require authentication for the server itself; account credentials are handled externally via environment variables.
