Agent payments, pay.sh and MPP

Let your agent choose its own paid tools, from providers you trust

pay.sh gives an agent a catalog of paid APIs, and the Machine Payments Protocol lets it pay for each call or open a session with a spending cap. That saves your team from wiring up every provider by hand. VeriProof makes sure the agent picks verified providers and stays under budget, and asks the budget owner when it wants to go further.

A research agent hits its cap

A pricing team's research agent needs competitor data every morning. It finds three providers in the pay.sh catalog and opens a payment session capped at $50 with the one it picks.

On Monday the agent reaches the cap by 10 a.m. and wants to open a second session. Your rule says a second session in one day needs the budget owner, so the request goes to her in the VeriProof Inbox with the provider, the spend so far, and what the agent was trying to finish. She approves it and notes why. The agent picks up where it stopped.

On Tuesday the agent finds a cheaper provider whose catalog entry doesn't match its published fingerprint. Your rule denies that payment before it goes out, and the agent falls back to the provider it used before.

What your rules can look at

Scroll horizontally to see all columns.

QuestionWhat the agent reports
Did the provider come from the pay.sh catalog, or from a URL the agent found elsewhere?Where the agent discovered the endpoint
Does the provider's catalog entry check out?Listed, verified, mismatched, or not listed
Is this one charge, or a session that keeps paying?Single charge, or a capped session and how much of the cap is left
Who approved the spend?A person in the pay CLI or MCP client, or auto-pay within a cap
Is this real money?A sandbox wallet or a live one

The VeriProof core SDK carries these as named fields, so a rule written for one agent reads the same way for the next. Your application wraps the payment call in the SDK's policy check, and pay.sh and your wallet keep handling the payment itself.

Questions from the teams involved

No. The request reaches them in Slack, Teams, email, or the VeriProof Inbox, with reminders, and escalates if it sits too long. In Slack or Teams they can approve or reject a routine request without leaving the message.

A governance engineer edits the rule and can run it in Shadow first. For a production agent, an administrator or compliance officer approves the change before it applies.

No. The core SDK handles it. Developers can start with the SDK quickstart, then get packages from Tools and SDKs in the customer portal.

Other ways agents pay: x402, AP2 mandates, and stablecoin supplier payments.