Apple Search Ads MCP API

The MCP server at searchad.ai/mcp exposes the same tool surface our own chat uses, to any MCP-compatible client. Plug Claude Desktop, Cursor, Cline, Continue, or your in-house agent into your Apple Search Ads account through SearchAd AI. The safety primitives travel with you.

Read the MCP docs Get a key

CLAUDE DESKTOP CONFIG

{
  "mcpServers": {
    "searchad": {
      "url": "https://searchad.ai/mcp",
      "headers": {
        "Authorization": "Bearer sak_live_..."
      }
    }
  }
}

Paste the key from Settings → API access. Restart your client. Your agent can read your account immediately; writes wait for the envelope or your tap.

Safety primitives

Five things that make autopilot safe.

Scoped keys

Each key carries a scope: read-only, read + write with envelope, or full write requiring per-call approval. Scope cannot be elevated client-side. Issue a new key for a new scope.

Daily spend envelopes

An envelope is a cap on daily Apple spend the agent can authorize on its own. Once breached, further writes require out-of-band approval (a notification you tap). You can set as many envelopes as you want, scoped per app or per market.

Out-of-band approval

When the agent wants to do something outside the envelope, SearchAd AI fires you a notification with the exact arguments. You tap Approve once; the agent gets a one-shot approval token bound to those arguments. No token, no write.

One-click kill switch

Settings → API access → Block agents. Revokes every key, blocks future calls, returns AGENTS_BLOCKED on every request until you unblock. A button you can hit from a phone in a meeting.

Full audit log

Every MCP call, with arguments and result, lands in your private audit log. Settings → API access → View agent activity. Filter by key, by tool, by time window. Forever, until you delete.

What an agent can actually do

The same tool surface, your own model running it.

The MCP server exposes the same tools SearchAd AI's own chat uses: list_my_apps, list_existing_campaigns, propose_campaign_plan, list_keyword_performance, deploy_campaign_plan, and the rest. Your agent reads your account, drafts plans into your workspace, and (within its envelope) deploys them. Everything is logged. Everything is reversible.

Limits to keep in mind: 60 requests per minute per key, 10,000 per day, 20 active keys per account, 365-day maximum key expiry (90-day default), 20 active spending envelopes per account. The MCP docs cover error codes, retries, and the raw HTTP shapes if you need them.

Read the full MCP documentation →

FAQ

Before you wire up autopilot.

What is MCP?+

Model Context Protocol is an open standard that lets AI clients (Claude Desktop, Cursor, Cline, Continue, others) talk to external tool servers over a structured JSON-RPC interface. SearchAd AI exposes an MCP server at /mcp so any compatible client can drive your Apple Search Ads account through chat.

Which clients does it work with?+

Claude Desktop, Cursor, Cline, Continue, and any other MCP-compatible client. Connection is a URL plus a bearer token; nothing to install on your side.

What stops an agent from spending my whole budget?+

Spending envelopes. Each key has a daily cap; actions that would exceed it require an out-of-band approval from you. Above that, a one-click kill switch revokes every key and blocks future calls instantly.

Does the MCP API skip the approval gate?+

Same gate, expressed differently. Spend up to the envelope is pre-authorized when you set the envelope. Spend above it needs a fresh approval token tied to specific arguments. The gate cannot be bypassed by prompt.

What about cost?+

MCP calls do not cost SearchAd credits. Your agent's LLM (Claude, GPT, or whatever you run) bills against your own subscription. Your SearchAd plan covers the tool surface and the underlying Apple API costs.