SpendGate
For agents that can pay

Check the spend before the agent pays.

SpendGate scores a planned payment and returns a verdict: approve, ask a human, or block. It does not hold funds and it does not pay the merchant. Your agent still decides whether to send the money.

approvePay. The plan fits policy.
ask_humanStop. A person should confirm.
blockDo not pay.

Try a score

This uses the free tier. Nothing is charged from the browser.

Example
53
ask_human

Do not pay yet. Ask a human to confirm this spend.

  • A $25 SaaS charge at a known API is inside the middle band.
  • Submit the form to replace this example with a live verdict.

How v1 decides

Higher spend, lower score

The amount penalty is min(75, round(25 × log10(1 + usd))). A missing or unknown merchant lowers the score further. Known hosts are listed on /v1/policy.

Category risk

crypto_transfer, irreversible, and gift_card are high. api_call, saas, and infra are medium. status_check is low. Anything else is unknown.

Two hard rules

Over a budget you pass in, the verdict is block. Strictly under $1, it is approve, unless that budget rule fires first.

Call it from an agent

curl -s $ORIGIN/v1/score \
  -H 'content-type: application/json' \
  -d '{"amount":12,"currency":"USDC","category":"api_call","merchant":"https://api.openai.com","agent_id":"planner"}'

MCP tools score_spend, get_quota, and get_api_key live at /.well-known/mcp. Agent notes are in llms.txt.