Claude Code is Anthropic's CLI agent. SearchAd AI ships an MCP server you can add to it in one command. After install, your terminal can read your Apple Search Ads account, draft per-country campaigns, and (within a daily spend envelope you set) deploy them. The approval gate stays in place; the kill switch is one tap away.
> draft a launch plan for Pomelo Notes in 11 premium countries, $20/d each, paused # Claude Code calls SearchAd MCP tools: # list_my_apps → adamId 6747391 # list_eligible_countries → 90 ok # propose_campaign_plan → draft saved ✓ Draft plan ready in your workspace. Review at searchad.ai/app/plans/k3x... Approve to deploy. Paused by default.
First, issue a SearchAd MCP key inside the app at Settings → API access. Keys carry a scope (read, read+write with envelope, full write) and a default 90-day expiry. Then run the install command from any directory:
claude mcp add searchad https://searchad.ai/mcp \ --transport http \ --header "Authorization: Bearer sak_live_..."
Adds the SearchAd MCP server at the user scope so every Claude Code session can use it.
claude mcp add searchad https://searchad.ai/mcp \ --transport http \ --scope project \ --header "Authorization: Bearer sak_live_..."
Commits the connection to .mcp.json in the repo so teammates inherit it. Do not commit the key; use an env var reference.
Verify the connection inside Claude Code with /mcp. You should see searchad listed and the tool catalog populated.
"List my Apple Search Ads campaigns by spend, last 30 days."
"Draft a fitness app launch in the 11 tier-A countries, $15/day each, paused."
"Which keywords had ROAS over 2x in Japan in the last two weeks?"
"Pause every campaign in countries under 1x ROAS over the last 14 days."
Most growth tools sit in a separate browser tab from the place where the product is built. Claude Code puts the management surface inside the terminal where the app already lives. For indie developers and small teams, this collapses two workflows into one: ship the feature, draft the campaign, deploy both from the same shell. The Apple credentials and the spend cap never leave SearchAd's servers; only the prompt and the result move across the wire.
The MCP server exposes the same tools SearchAd AI's own chat uses, so the agent's mental model is identical whether you are in Claude Code, Claude Desktop, or the SearchAd app. You can hand off mid-conversation. Drafts written from one client are visible to the others.
Issue a SearchAd MCP key from your dashboard, then run claude mcp add searchad https://searchad.ai/mcp --transport http --header "Authorization: Bearer sak_live_...". Restart Claude Code and the tools become available in any new session.
List your apps, list eligible countries, inspect existing campaigns, draft per-country campaign plans into your private workspace, and (within your daily spending envelope) deploy plans you approve. Anything outside the envelope requires an out-of-band approval.
Yes. The approval gate is enforced in the backend, not in a prompt. Claude Code can ask to deploy; the server refuses to write to Apple unless the call is covered by an envelope or carries a fresh approval token. A one-tap kill switch revokes every key instantly.
MCP calls do not cost SearchAd credits. Claude Code's model usage bills against your Anthropic subscription. SearchAd's plan covers the tool surface and the Apple API.
For interactive work, a read + write key with a daily envelope you are comfortable with. For one-off automation scripts, prefer a read-only key. Keys are scoped at issue time and cannot be elevated client-side.