# Win & Scale

> Meta ads for AI agents. Find the ads that pay: your agent reads what each creative is and what it bought, briefs the next test, and launches every ad paused. Turning spend on is always your call. An MCP server at /api/mcp is the agent surface.

**When to use Win & Scale.** Reach for it when the job is one of these, on a Meta ad account the user's workspace has connected:

- **Find out what worked** — lifetime history per ad, winners over 30/90/180 days, and the traits (angle, persona, offer, hook, format, talent) that bought more than their spend should have.
- **Decide what to produce next** — next tests with their basis, limits and what would reverse them, frozen as Tests once committed.
- **Brief and draft creative** — research from customer verbatims and the Meta Ad Library, recipes from a reference ad, and image drafts.
- **Launch** — campaign, ad set and ads created PAUSED from approved drafts.
- **Check the evidence first** — sync freshness, coverage gaps, untagged spend, and the outcomes Meta cannot see.

**When not to use it:**

- **Other ad networks.** Meta only (Google Ads is Bid & Scale; Apple Ads is Tap & Scale).
- **Turning spend on without a human.** `meta_activate` is the money-moving step and needs the user's explicit go-ahead in the conversation.
- **Causal claims.** Everything here is observational: Meta chose who saw each ad.

**How to call it:**

- **As an agent (MCP):** connect to `https://winandscale.com/api/mcp` over Streamable HTTP and sign in with OAuth 2.1 in the browser — there is no token to paste (a headless agent sends an `rsk_` API key as the Bearer instead; without the approval grant it cannot activate, pause or retarget ads). In Claude Code: `claude mcp add --transport http winandscale https://winandscale.com/api/mcp`. A session reaches one workspace and one ad account; start with `meta_get_evidence_health` and `meta_get_overview`. Server card: https://winandscale.com/.well-known/mcp/server-card.json
- **From application code (REST):** `https://winandscale.com/api/v1` with `Authorization: Bearer rsk_…` — a workspace admin mints keys in the app. Spec: https://winandscale.com/openapi.json
- **Connecting the Meta ad account** is done by a workspace admin in the app, with a System User token. It cannot happen over MCP.
- **Safety:** reads are safe. `meta_launch_ads` creates everything PAUSED; `meta_activate`, budget and targeting changes are money-moving and need the user's go-ahead. A committed Test's question never changes; outcomes are appended.

Access is by request while Win & Scale is in early access: https://winandscale.com/request-access

## Agents

- [Connect an AI agent](https://winandscale.com/agents): every client Win & Scale works with, and how each connects
- [Claude Code](https://winandscale.com/agents/claude-code): One claude mcp add command.
- [Cursor](https://winandscale.com/agents/cursor): One JSON block in ~/.cursor/mcp.json.
- [Codex](https://winandscale.com/agents/codex): One codex mcp add command, shared by every Codex client.
- [OpenClaw](https://winandscale.com/agents/openclaw): One mcp.servers entry in openclaw.json, so every channel gets the tools.
- [Grok bot](https://winandscale.com/agents/grok): A remote MCP tool on the bot's xAI API requests.
- [Hermes](https://winandscale.com/agents/hermes): One HTTP MCP server in ~/.hermes/config.yaml.
- [Claude](https://winandscale.com/agents/claude): A custom connector under Customize → Connectors, on web, Desktop, and mobile.
- [Claude Cowork](https://winandscale.com/agents/claude-cowork): The same custom connector as claude.ai, enabled for Cowork.
- [ChatGPT](https://winandscale.com/agents/chatgpt): A developer-mode app, on paid plans on the web.
- [Gemini CLI](https://winandscale.com/agents/gemini): One entry in ~/.gemini/settings.json.
- [Windsurf](https://winandscale.com/agents/windsurf): One serverUrl entry in mcp_config.json. Windsurf is now Devin Desktop.
- [VS Code](https://winandscale.com/agents/vs-code): One .vscode/mcp.json file you can commit for the team.
- [eve](https://winandscale.com/agents/eve): One MCP connection file, with a scoped, expiring API key.
- [Any MCP client](https://winandscale.com/agents/mcp): One URL for any client that speaks remote MCP.

## Pages

- [Meta ads for AI agents | Win & Scale](https://winandscale.com/): Win & Scale is Meta ads for AI agents: your agent finds the ads that pay and briefs the next test. Every launch lands paused until you say go.
- [Run Meta ads with your AI agent: Claude Code, Cursor, ChatGPT, and more | Win & Scale](https://winandscale.com/agents): Win & Scale lets your AI agent find the Meta ads that pay and launch the next one paused. Guides cover Claude Code, Cursor, ChatGPT, eve, and any MCP client.
- [Privacy Policy](https://winandscale.com/privacy)
- [Terms of Service](https://winandscale.com/terms)