Skip to main content
Product/MCP Server
Live

11 tools. One protocol. Real payments.

Connect any AI agent to Ovra's payment infrastructure via Model Context Protocol. No SDK, no npm packages — agents connect through MCP tools and transact autonomously.

Connect in seconds

Install the Ovra Agent Skill or add the MCP config directly.

Agent Skill (recommended)
$ npx skills add Ovra-Labs/ovra-pay

One command. Auto-configures tools, auth, and server connection.

mcp.json
{
  "mcpServers": {
    "ovra": {
      "command": "npx",
      "args": ["-y", "@ovra/mcp@latest"],
      "env": {
        "OVRA_API_KEY": "sk_..."
      }
    }
  }
}

Drop into any MCP-compatible client config. Get your key →

Available Tools

11 MCP tools covering the full payment lifecycle.

01
ovra_intent

Declare and manage payment intents under your spending policy

02
ovra_card

Issue and manage virtual cards (single-use or reusable) per agent

03
ovra_search

Browse the 10-vertical catalog and per-vertical query schemas before composing a buy

04
ovra_buy

Canonical AP2 buy — mints DPAN, signs the 4-mandate chain, debits wallet, captures with the merchant

05
ovra_order

Look up, list, verify and refund agentic-commerce orders (UCP envelope + AP2 chain)

06
ovra_transaction

List and look up payment history (read-only)

07
ovra_account

List and get accounts; funding and wallet creation are done in the dashboard

08
ovra_payment

Send money — account-to-account or account-to-vendor SEPA

09
ovra_agent

Provision and manage AI agents with scoped tokens and policy binding

10
ovra_customer

Read the workspace customer envelope (profile + GDPR moved to the dashboard)

11
ovra_me

Identity probe — who the current API key / agent token resolves to

How it works

From intent to payment in three steps.

01

Declare intent

ovra_intent({
  action: "declare",
  agentId: "ag_...",
  expectedMerchant: "aws.amazon.com",
  expectedAmount: 100, // decimal euros
  purpose: "EC2 reserved instance"
})

The agent declares what it wants to buy and why. The policy engine evaluates the request before any payment instrument is issued.

02

Approve and buy

ovra_intent({ action: "approve", intentId: "in_..." })
ovra_buy({ action: "buy", offerId: "prd_notion-business" })
// → { order_id: "cmo_...", dpan: "4***7823", exp: "08/27" }

The intent is approved (by policy or human). The AP2 buy pipeline mints a tokenized DPAN, signs the 4-mandate chain, debits the wallet, and captures with the merchant. The agent never sees real card data.

03

Pay and verify

// Agent fills checkout with DPAN
// Ovra verifies charge against intent
// → transaction_id: "txn_..."

The agent completes checkout. Ovra matches the network charge to the approved intent and records the full audit trail.

Compatible with any MCP client

Works with Claude, Cursor, GPT, and any MCP-compatible agent framework. No vendor lock-in.

Claude DesktopCursorWindsurfGPT (via MCP bridge)Any MCP Client

Design principles

Security architecture, not security theater.

Intent-first architecture

Agents declare purchases before instruments are issued. No credential without an approved intent.

Zero-knowledge credentials

Agents never see real card data. Every transaction uses a network token with a single-use cryptogram.

Controls in the transaction path

Policies are enforced at the API layer, not in prompts. Blocked transactions never reach the card network.

Read-only policies

Agents cannot modify, disable, or escalate their own spending rules. Only human users can change policies.

Ready to connect your agent?

Get API access, connect via MCP, and let your AI agents transact autonomously.