AI ads infrastructure
Which AI Tools Work With Google Ads?
The AI tools that connect to Google Ads in 2026 — Claude, ChatGPT, Gemini, Cursor, and Windsurf — plus the MCP setup and eligibility limits for each.
Any AI tool that supports remote MCP (Model Context Protocol) can connect to Google Ads through a server such as NotFair. Current options include Claude, ChatGPT/Codex, Gemini CLI, eligible Gemini web-app users through custom apps, Cursor, Windsurf, and Cline. The exact setup and feature availability still vary by client.
AI tools that work with Google Ads
- Claude Desktop — fully supported via the Connectors UI. One-click install through notfair.co. Best for marketers and agency operators.
- Claude Web (claude.ai) — supported via the same hosted connector. No local install required.
- Claude Cowork — supported. Same connector setup as Desktop.
- Claude Code — supported via the AdsAgent plugin (claude plugin install nowork-studio/adsagent-plugin). Best for developer/agency workflows.
- ChatGPT — supported via Codex connectors. Same NotFair auth works simultaneously with Claude.
- OpenAI Codex CLI — supported. Add the NotFair MCP endpoint to ~/.codex/config.toml.
- Gemini CLI — supported through its MCP configuration. Add the hosted NotFair endpoint and complete OAuth.
- Gemini web app — custom MCP apps are supported for eligible personal-account users in the United States, in English, inside Spark tasks. Availability is narrower than Claude or the CLI.
- Cursor — supported via stdio MCP config in ~/.cursor/mcp.json.
- Windsurf — supported via the same stdio MCP config pattern.
- Cline — supported. Add NotFair to the MCP servers list in Cline's settings.
AI tools with limited or no direct MCP support
- Gemini web app outside the eligible custom-app rollout — MCP app availability depends on account type, country, language, and Spark-task access.
- Microsoft Copilot — no MCP client support. Has its own plugin model.
- Perplexity — no MCP client support. Browse-and-summarize only.
- Most chatbot builders (e.g., HubSpot AI) — closed integration models, no MCP.
Client support changes quickly. Check the client vendor's current MCP documentation before choosing a production workflow.
Which one should you start with?
- Marketers, agency owners — Claude Desktop. The Connectors UI is the lowest-friction setup.
- Developers, technical teams — Claude Code or Cursor. Faster iteration, scriptable, integrates with the rest of your dev workflow.
- ChatGPT-first teams — ChatGPT via Codex connectors. Same NotFair auth, no need to switch tools.
- Gemini-first technical teams — Gemini CLI. Use the web app only if your account has custom MCP apps in Spark.
- Mixed teams — connect both Claude and ChatGPT to the same NotFair account. Compare reasoning across models on the same data.
What about Google's own AI features?
Smart Bidding, Performance Max, and the AI features inside Google Ads are different — they run inside Google's platform and optimize within their narrow scope. They don't replace external AI assistants like Claude. They work alongside: use Smart Bidding for bid execution, use Claude or ChatGPT (over MCP) for diagnosis, recommendations, and oversight.
FAQ
Move from article to live account work
Connect your account first, then use an MCP client to turn the article workflow into a reviewed account action.