Run your Google Ads and Meta accounts from inside Claude.
Add the SpendSignoff remote MCP server to Claude and ask, in plain English, what to change. Claude reads live and stages a draft — you approve the spend.
Our Q2 Brand campaign on Google Ads is overspending. Move 20% of its daily budget to the Prospecting campaign and draft the change.A draft moving $80/day from Brand to Prospecting — projected pacing back inside the daily envelope, staged for your approval.
Staged as a draft. Nothing reaches your ad accounts until you approve it.
Claude Desktop and the Claude apps speak MCP natively. You point them at the SpendSignoff server once, sign in over OAuth, and your connected ad accounts show up as tools Claude can call.
Claude is issued mcp.read and mcp.draft scopes only. It can pull live numbers and write a proposed change, but pushing money live is a separate step you take in the SpendSignoff dashboard.
What it does
Add the server in Settings → Developer
Open Settings → Developer → Edit Config, add the SpendSignoff entry pointing at https://mcp.spendsignoff.com/mcp, and restart Claude. The OAuth sign-in runs on first use.
OAuth in the browser, not a pasted key
Claude opens the SpendSignoff consent screen in your browser. You grant mcp.read and mcp.draft — there is no mcp.approve scope to grant, by design.
Ask in the chat, read comes back live
Type a question about a campaign and Claude calls the read tools directly: spend, ROAS, pacing, disapprovals, and budget headroom, pulled fresh from Google Ads or Meta.
Drafts land in the approval queue
When you ask for a change, Claude stages a typed draft. It appears in your SpendSignoff approval queue with the full before → after diff — it does not go live from the chat.
How it works
Read
Ask Claude about an account. It calls the read tools and answers from live Google Ads and Meta data — no spend touched.
Draft
Tell Claude what to change. It writes a typed draft (a budget shift, a bid change, a pause) and stages it for review.
Approve
You open the SpendSignoff queue, read the diff, and use the two-step approve to push it live. Every action is signed into the audit log.
Questions
Can Claude change my budget on its own?
No. Claude only ever holds read and draft scopes. It writes the draft; pushing it live is a separate two-step approval you take in the SpendSignoff dashboard.
Which Claude apps support this?
Any Claude client that supports remote MCP servers — Claude Desktop and the Claude apps. You add the SpendSignoff server once in Settings → Developer.
Do I paste an API key?
No key. Connection is OAuth 2.1: Claude opens the SpendSignoff consent screen in your browser and you grant mcp.read and mcp.draft.
Connect it to
Put Claude on autopilot — with your hand on the switch.
Free to connect and read. You only pay when you want the operator drafting around the clock.