>_

SpendSignoff documentation

The always-on operator for your ad accounts. Read and draft from your AI client — approve before anything goes live.

Knowledge base

Using SpendSignoff from multiple AI clients

One SpendSignoff account, multiple AI clients connecting to the same MCP server. The session model, draft ownership, and audit log are designed for this. This page explains what changes and what stays the same.

Your AI can read and draft — it can never spend without your approval.

One MCP endpoint, many clients

Every AI client connects to https://mcp.spendsignoff.com/mcp over Streamable HTTP + OAuth 2.1. The server is stateless between requests. What differs per client is the OAuth token it holds and the session it opens.

Claude Desktop, ChatGPT, and Cursor are live today. Claude Code, Codex, Windsurf, and Gemini CLI are coming soon. Each one goes through the same OAuth flow and receives mcp.read and mcp.draft scopes — nothing more.

What is shared across clients

  • Account data — reads pull from the same synced account state regardless of which client queries.
  • Approval queue — drafts created by any client appear in the same SpendSignoff approval queue. You review and approve in the app, not the AI client.
  • Audit log — every approved action is recorded with the source session. You can see which client initiated the draft that led to an approved change.
  • Safety model — mcp.approve does not exist on any client. Approval always happens in the SpendSignoff app.

What is not shared

  • Conversation context — each AI client has its own conversation memory. A draft created in Claude Desktop is visible in the approval queue, but ChatGPT does not know the conversation that produced it.
  • OAuth tokens — each client holds its own token. Revoking one client does not revoke others.
  • Active session — a Streamable HTTP session started by Claude Desktop and one started by Cursor are separate; there is no cross-client state in the MCP server.

Draft contention between clients

Two clients can draft changes for the same account in the same window. The approval queue shows all pending drafts. If two drafts target the same field on the same campaign, approving the first one marks the second STALE — the account state changed, so the second draft is no longer safe to apply.

This is the correct behavior. It is better to discard a stale draft and re-draft than to apply a change against outdated numbers.

The approval queue is the integration point

Multiple AI clients produce drafts; a human reviews and approves them in one place. The queue is the single surface where all client activity converges. You do not need to track which client produced which draft — the audit log records that for you.

Next

Compatibility

Supported AI clients today and what each one can do.

    Using SpendSignoff from multiple AI clients — SpendSignoff Docs