PAYING.CO GUIDES
MCP for payments, explained.
What the Model Context Protocol means for payments, what a payments MCP server can safely do, and the design rules that keep card data and money movement under control.
4 min read · By Kevin Medina · Updated November 19, 2026
The short version
MCP lets AI assistants like Claude and ChatGPT call payment tools such as creating a payment link or checking a payment status. Start with read and request actions, keep card data out of the model through hosted payment pages, and put limits, human approval and audit logs around anything that moves money.
What MCP is
The Model Context Protocol (MCP) is an open standard that lets AI assistants connect to outside tools and data. A service publishes an MCP server that lists the actions it allows, and an assistant such as Claude or ChatGPT can call those actions on a user's behalf. For payments, that means an assistant can do things like create a payment request or check whether one was paid.
What a payments MCP server can do
| Level | Example | Risk |
| Read | Look up a transaction or a payment link's status | Low, if access is scoped |
| Request | Create a payment link or a request to pay that the customer completes on a hosted page | Moderate, and card data never touches the assistant |
| Execute | Move money or charge a card directly | High, needs hard limits and human approval |
Most teams should start at the first two levels. A hosted payment link keeps card numbers out of the conversation and out of your PCI scope.
Design rules for an MCP payments server
- Least privilege: scope each key or token to the few actions it needs, and separate sandbox from live.
- No card data in the model: send the customer to a hosted page or a tokenized flow instead.
- Human in the loop: require a person to confirm anything that creates a charge or sends money.
- Limits: cap amounts, rate and recipients, and fail closed.
- Idempotency: a retried tool call must not create a second payment.
- Audit: log every call with who or what made it, and the result.
An example: AskPay
AskPay is our payment-links platform. It creates links customers pay by QR code, email, SMS or button, and it exposes a REST API, webhooks, a sandbox and a native MCP server for Claude and ChatGPT. An assistant connected to it can create a payment request without ever handling a card number. See the AskPay developer page and the API and MCP guide.
MCP and agent-initiated checkout are different things
MCP lets a person's assistant work with your payment tools. Agent-initiated checkout is a separate question: how a merchant accepts a purchase an AI agent makes for a customer, through card network programs and emerging protocols. Both matter for platforms and ISOs. We cover the second in our page on agentic payments for ISVs and ISOs.
Frequently asked questions
What is MCP in payments?
MCP, the Model Context Protocol, is an open standard that lets AI assistants call outside tools. In payments, an MCP server lets an assistant do things like create a payment request or check a payment status, within the limits the server enforces.
Is it safe to let an AI assistant take payments?
It can be, if the design is careful: scoped credentials, no card data passing through the model, caps on amounts, human approval for anything that moves money, and a full audit trail. Starting with read and request actions is safer than direct execution.
What is an MCP server for payments?
It is a service that publishes a list of payment actions an AI assistant is allowed to call, such as creating a payment link or looking up a transaction, along with the rules for using them.
Does Paying.co have an MCP payments server?
Yes. AskPay, our payment-links platform, includes a native MCP server for Claude and ChatGPT alongside its REST API. Details are in the API and MCP guide at ask.paying.co/mcp.
Does MCP bring card data into PCI scope?
It can if card numbers pass through the assistant or your server. A design that sends the customer to a hosted payment page keeps card data out of the conversation. Confirm your own scope with your assessor.
What is the difference between MCP and agentic payments?
MCP connects an assistant to your tools. Agentic payments, in the narrower sense, means accepting a purchase that an AI agent makes for a customer. They overlap, but they are different projects.
Want an AI assistant to create payment requests safely?
Paying.co builds MCP-native APIs and payment tools. Tell us what you want an assistant to do, and we will scope where the limits and approvals belong.
Talk to us Book a meeting with a founder