Spend Control for OpenAI Codex
Codex is OpenAI's autonomous coding agent — and it ships with the ability to act. Acting includes spending. A guard tool makes the spending policy real.
Why Codex agents overspend
Long autonomous runs call tools thousands of times.
Paid tool calls and purchases compound during agent loops.
OpenAI's own usage limits only cover OpenAI.
How it works
Add sipi.bot as an MCP tool. Before any spend — compute, API credits, payments — Codex calls the guard and gets APPROVED, BLOCKED, or FLAGGED.
Rules that fit Codex workflows
Per-transaction cap for compute purchases.
Merchant allowlist for known vendors.
Velocity limit so retry loops die in seconds.
MCP registration
{
"mcpServers": {
"sipi": {
"command": "npx",
"args": ["-y", "@sipi/bot-mcp"]
}
}
}Codex discovers the tool on startup.
FAQ
Does this replace OpenAI's own limits?
No — keep them for OpenAI. sipi.bot covers every merchant, including purchases outside OpenAI.
Is the decision model-free?
Yes — deterministic rules only, ~5 ms, nothing to inject.
What happens on BLOCKED?
The agent stops, the money doesn't move, and the attempt is logged.
Related
Stop the next $12,400 night.
One API call (or MCP tool) in front of every agent transaction — APPROVED, BLOCKED, or FLAGGED, deterministic, ~5 ms, fully logged.
See plans — from $99/mo Try a live check