Set up a paid-media AI agent with OpenAI Codex and SpendSignoff
OpenAI Codex runs in a sandboxed environment with tool access. Once you add SpendSignoff as an MCP tool, it gains read and draft access to your ad accounts. The setup is three steps: config, OAuth, first query. The rest is prompting.
1. Create your SpendSignoff account
Sign up at spendsignoff.com. The free tier supports one connected ad account and a working set of read and draft calls — enough to test the full workflow before committing to a paid plan.
2. Add SpendSignoff to Codex tool config
In your Codex project or agent configuration, add SpendSignoff as an MCP tool endpoint:
Codex tool config
{
"tools": [
{
"type": "mcp",
"name": "spendsignoff",
"url": "https://mcp.spendsignoff.com/mcp"
}
]
}Codex sandboxed environments
mcp.spendsignoff.com is allowed. SpendSignoff only uses port 443.3. Connect Google Ads via OAuth
Start a Codex session and ask it to connect your Google Ads account:
Codex prompt
Connect my Google Ads account using SpendSignoff.4. Connect Meta Ads
Repeat for Meta:
Codex prompt
Connect my Meta Ads account using SpendSignoff.5. Run a cross-platform query
You can connect both platforms in the same session or come back to Meta later. After both are connected, cross-platform queries work:
Codex prompt
Compare my Google Ads and Meta Ads ROAS for the last 30 days.
Which platform is delivering better returns for my prospecting campaigns?5. Draft and approve your first change
Ask Codex to propose an action based on what it finds:
- Codex drafts three budget changes — one per ad set — via SpendSignoff.
- Each draft lands in your SpendSignoff Approval Queue as a before/after diff.
- Open the dashboard, review each draft individually, and approve the ones you agree with.
- Codex cannot execute the changes. The approval step is always yours.
Codex prompt
Find the three lowest-ROAS ad sets across both platforms this month.
Draft a budget reduction of 30% on each one.Cross-platform approval works the same way
FAQ
- Is Codex a good choice for ad management compared to Claude or ChatGPT?
- Codex performs well for structured operations — bulk edits, data extraction, templated campaign creation. For open-ended diagnosis or nuanced performance analysis, Claude with a marketing context tends to give more actionable output. The good news is that the SpendSignoff connection works identically for all three, so you can switch based on the task.
- Can I automate Codex to run ad checks on a schedule?
- Yes, if you run Codex programmatically via the API. You can schedule a Codex agent to query account performance and push drafts to the approval queue on a regular cadence. Approvals still require a human — you get a notification when drafts are waiting.
- What happens if Codex loses the OAuth session mid-task?
- SpendSignoff issues long-lived encrypted tokens. If the token expires, the next read or draft attempt returns an auth error, and Codex will prompt you to re-authorize. Active tasks are not lost — they simply pause until the connection is restored.
Connect an account read-only and watch the operator work.
Reads are free on every plan. Nothing spends without your two-step approval.
Related reading