MCP ServerOfficialHTTP

PayPal MCP Server

PayPal ships an official hosted MCP server covering invoices, orders, refunds, subscriptions, disputes and transaction reporting. Any MCP client can raise an invoice, capture a payment or investigate a dispute — and there is a separate sandbox endpoint, so you can build the whole workflow without touching real money.

Open Agent Studio PayPal MCP server quickstart

Hosted URL

https://mcp.paypal.com/mcp

Suggested model

Claude Sonnet 5

Chat with 60+ AI models on the same workflow — switch to a different model mid-conversation and re-run the same prompt, or use Compare mode to put several side-by-side and balance quality vs. cost.

Auth

OAuth through PayPal login on the remote server. Use the sandbox endpoint until the workflow is proven — these tools move real money.

What the PayPal MCP server does

How models use it and what it is built for.

The PayPal MCP server exposes PayPal's payment APIs as MCP tools, aimed squarely at agentic commerce. The invoicing surface is the most immediately useful: an agent can create an invoice, list existing ones with filters, and send reminders on the overdue ones. Orders and payments cover creating an order and capturing payment against it, plus refunds. Beyond that it reaches subscriptions, dispute management and transaction reporting, which together make it viable as the finance-operations end of an agent rather than only a checkout step. PayPal runs it as a remote server on their own domain over both streamable HTTP and SSE, and also publishes a local npx package for people who would rather hold their own credentials.

Tools the PayPal MCP server exposes

Typical tools an AI model can call. Exact names vary by version.

create_invoice — raise a new invoice
list_invoices — list invoices with pagination and filtering
send_invoice_reminder — chase an unpaid invoice
create_order / pay_order — create an order and capture payment against it
refunds — issue a refund against a captured payment
subscriptions — create and manage recurring billing
disputes — list and manage cases
transactions — reporting and reconciliation reads

Connecting to PayPal

Taken from the official PayPal documentation — see PayPal MCP server quickstart for the full reference.

Examples

Production (streamable HTTP)

https://mcp.paypal.com/mcp

Live account. Every call here has real financial consequences.

Sandbox

https://mcp.sandbox.paypal.com/mcp

Build and test the whole workflow here first. There is no undo on a captured payment.

Production (SSE)

https://mcp.paypal.com/sse

Legacy transport, for clients that do not yet speak streamable HTTP.

Client configuration

Claude Code — remote server over SSE

Authenticates through PayPal login (OAuth) on first connection.

claude mcp add --transport sse paypal https://mcp.paypal.com/sse

Local npx package

For holding your own credentials rather than using the hosted server. Set PAYPAL_ENVIRONMENT to sandbox while testing.

{
  "mcpServers": {
    "paypal": {
      "command": "npx",
      "args": ["-y", "@paypal/mcp"],
      "env": {
        "PAYPAL_ACCESS_TOKEN": "<your-access-token>",
        "PAYPAL_ENVIRONMENT": "SANDBOX"
      }
    }
  }
}

Example prompts to try

Copy any of these into MCP Agent Studio after connecting.

List every invoice more than 30 days overdue and send a reminder on each.
Create an invoice for 2,400 USD to this customer with net-14 terms.
Summarise the open disputes and tell me which are close to their response deadline.
Reconcile last week's transactions and flag anything that looks like a duplicate charge.

Models on MCP Playground

This is not a single-model product: you get the same MCP connection with 60+ models (Claude, GPT, Gemini, DeepSeek, open-weight, and more), you can switch mid-conversation, and you can open Compare mode to run the same prompt against multiple models at once. The card above is a suggested starting point for this server — not the only choice.

Default pick for PayPal

Claude Sonnet 5

These tools move money. Sonnet 5 is the cheapest model here that reliably confirms intent and gets amounts and currency codes right rather than approximating them.

Open Compare mode (side-by-side)Try this server in Agent Studio

Check an AI agent can actually use the PayPal MCP server

Listing tools proves the server is reachable, not that a model can work with it. Evals go further: they read every tool on the server, write a test suite from its real schemas, and run it — code decides pass/fail on the responses (schema conformance, error codes, pagination, result caps) while a scoring model grades plain-English tasks driven through the tools.

Get a pass/fail report per tool with the evidence behind each verdict — and replay the same suite after every schema change. Destructive tools are excluded from the run.

Run evals

Try the PayPal MCP server in your browser

Open MCP Agent Studio with the connection pre-filled. Add your token, pick any of 60+ models, and start chatting — no install required.

Open Agent Studio

PayPal MCP server — FAQ

Common questions about connecting, scoping and using it safely.

Does PayPal have an official MCP server?

Yes. PayPal runs a hosted remote MCP server at https://mcp.paypal.com, documented on PayPal's own developer site, supporting both streamable HTTP and SSE. They also publish a local npx package, @paypal/mcp, for teams that would rather hold their own credentials than use the hosted endpoint.

Is there a sandbox I can test against?

Yes, at https://mcp.sandbox.paypal.com/mcp, and you should use it. This is the one server on this site where a mistaken tool call moves real money — an agent that misreads an amount or captures the wrong order cannot be undone with a retry. Build the entire workflow in sandbox and only switch the URL once it is boring.

What can the PayPal MCP server actually do?

Invoicing (create, list with filters, send reminders), orders and payments (create an order, capture payment, issue refunds), subscriptions, dispute management, and transaction reporting. The invoicing and reconciliation tools tend to deliver value fastest, because they are high-volume, low-judgement work that still needs a human to approve the outcome.

Is it safe to give an AI agent access to payments?

Treat it as you would a junior employee with your card: scope tightly, test in sandbox, and keep a human in the approval loop for anything that captures or refunds. The genuine risk is not malice but confident error — a model that misparses a currency or an amount. Watching every call with its full arguments in MCP Agent Studio before trusting a workflow is the cheapest safeguard available.

How is this different from the Stripe MCP server?

They cover the same conceptual ground — payments, invoices, subscriptions, refunds — for different processors, so the right one is simply whichever holds your money. PayPal's surface leans harder into invoicing and dispute handling, which reflects where PayPal is strongest commercially. If you run both processors, you can connect both servers to the same client and let the agent reconcile across them.