Agent Spend Authorization: Check Before the Payment Call
If you are building an agent that spends money, the first design question is not how much it may spend. It is where the check happens. This post explains agent spend authorization: a decision made before each payment or paid call, by a policy the agent cannot rewrite.
Why a cap alone is not enough
A cap bounds the damage. It does not stop the next call. If an agent loops, or is pushed by a bad prompt toward a merchant you never intended, a monthly cap only tells you after the fact that the budget is gone, and it usually cannot say which agent, merchant, or rule was involved. Keep caps as the outer wall. For the longer argument, see Hard Spend Caps Are Becoming the Default.
What pre-call authorization means
Before every payment, the agent (or the runtime orchestrating it) makes one /authorize call describing the spend: which agent, how much, which currency, which merchant. It pays only if the answer is yes. The decision and the rule behind it are logged.
POST /v1/integration/authorize
X-API-Key: <your key>
{ "agentId": "...", "amount": 42.50, "currency": "USD", "merchant": "..." }
Check the API reference for the exact request and response shape.
The three outcomes
- Approve. The request satisfies the policy. The agent proceeds.
- Deny. A rule failed, such as an amount over the per-transaction limit. The agent does not pay, and the reason is recorded.
- Route to human approval. The request is over an approval threshold. A person approves or denies it, and it expires if nobody answers.
The routing design is covered in Routing Agent Payments Between Auto-Approve, Human Review, and Deny.
What the policy covers
A policy is a set of rules evaluated on each request:
- Amount: a ceiling per transaction.
- Period: daily and other budgets, per agent.
- Merchant: an allowed list of merchants.
- MCC: allowed merchant category codes.
- Currency: which currencies the agent may spend in.
- Velocity: limits on how fast spend can accumulate.
- Business hours: windows when spend is allowed.
- Approval thresholds: amounts above which a human must decide.
Why enforcement sits outside the agent
A rule written into the prompt is a suggestion to the model. A rule evaluated in a separate service is a decision the model cannot argue with. Because the check runs outside the agent, a confused or manipulated agent can still ask, but it cannot grant itself an approval. The caveat is real: a check only covers calls that go through it, so your integration has to ask first.
Paitify is a policy and audit layer. It does not move, hold, or custody funds, and it does not sit between your agent and the vendor.
Using it from Claude or an MCP client
Paitify exposes an MCP endpoint so an MCP-capable client can call it directly. It offers five tools of its own: authorize_spend, capture_spend, cancel_spend, get_remaining_budget, and get_authorization_status. These are Paitify's own tools; it does not front third-party MCP servers. For the wider picture, see MCP Stateless Payments and Policy Enforcement.
Try it
Pick one agent, put its paid calls behind an authorize step, and read the decision log for a week. The free tier includes one agent, one policy, and 1,000 authorizations per month. Read how it works, or watch the demo on the homepage.