Capri AgentPay is a governed payment layer for AI agents that handles HTTP 402 challenges, enforces budgets and approval policies, and issues signed receipts — all without giving agents direct access to signing keys.
What is Capri AgentPay?
Capri AgentPay is the first product from Capri AI, purpose-built for teams shipping agent systems that need to pay APIs, internal tools, or infrastructure. It sits between the agent and any signer: the agent sends a typed fetch request with a policy name; if a 402 Payment Required challenge is returned, Capri evaluates the policy in a fixed order (merchant allowlist, per-transaction cap, daily budget, auto-approve threshold), optionally pauses for human review, settles the payment, retries the original request with proof of payment attached, and returns a signed receipt that can be verified offline. It runs on any HTTP‑capable agent stack and is currently available in a Base Sepolia sandbox for settlement.
Key Features
- x402 Handling — Automatic interception of HTTP 402 challenges from paid APIs and MCP tools; the agent only needs to call one typed surface.
- Policy‑gated payments — Four‑step policy evaluation: merchant allowlist, per‑transaction cap, daily budget, and an auto‑approve threshold that triggers human review when exceeded.
- Human‑in‑the‑loop approvals — Operators see exact amounts, assets, chains, and recipient addresses before approving or denying; policies can also hard‑deny outside defined bounds.
- Signed receipts — Every settled payment produces a receipt object with hashes, timestamps, and verification fields that can be validated offline.
- Six narrow MCP tools — Instead of raw key access, agents get exactly six purpose‑built MCP tools for payments, budgets, and approval workflows.
- Developer surfaces — VS Code extension for local debugging, a quickstart demo, and deployment guide for Pages.dev.
- Sandbox settlement — Early access runs on Base Sepolia testnet, allowing teams to test payment flows without mainnet funds.
Who is it for?
- Agent platform teams — Engineers who need to let agents call paid APIs (e.g., premium data feeds, inference endpoints) without handing out private keys or exposing wallets.
- Security and compliance officers — People who require audit trails, policy enforcement, and verifiable receipts before agents can spend money.
- Developers building agent toolchains — Teams integrating agentic workflows that touch internal billing systems or require approval gates for high‑value transactions.
What can you do with Capri AgentPay?
- Protect an agent’s payment path — Wrap any HTTP call in a
agentpay.fetch with a policy name; Capri handles 402s, policy checks, and receipts automatically.
- Set per‑agent budgets and caps — Define daily budgets and per‑transaction limits per agent, with optional merchant allowlists to lock down which endpoints can be paid.
- Add human review for expensive requests — Configure a threshold above which every payment request goes to an operator review queue with full context before settlement.
How does Capri AgentPay work?
- The agent calls
agentpay.fetch(url, { policy: "research-agent" }).
- If the endpoint responds with a 402 Payment Required challenge, Capri intercepts the request.
- The policy engine evaluates in order: merchant allowlist, per‑transaction cap, daily budget, auto‑approve threshold.
- If any check fails or human review is required, the operator sees the exact amount, asset, chain, and recipient for approval.
- Once approved, Capri settles the payment, stores a signed receipt, and retries the original request with proof attached.
