Can AI agents automate PayPal?

PayPal exposes a mature JSON REST platform for orders, authorizations, captures, refunds, invoicing, payouts, subscriptions, catalog products, transaction search, disputes, merchant onboarding, tracking, identity, and webhook administration. Sandbox and live applications exchange a client ID and secret for OAuth 2.0 bearer tokens; production use requires a PayPal Business account and features, scopes, and availability differ by market. PayPal publishes OpenAPI descriptions and current server SDKs for Java, .NET, PHP, Python, Ruby, and TypeScript, although the newer SDK layer covers fewer endpoints than the underlying REST surface. Its open-source first-party MCP server runs locally through `@paypal/mcp` and handles useful reads and writes across invoices, orders, refunds, disputes, shipment tracking, products, subscriptions, and transaction reporting. The MCP client supplies a short-lived PayPal access token and can expose money-moving or externally visible operations, so secret rotation, sandbox-first tests, least tool selection, and approval gates are essential. Merchant fees, regional eligibility, account capabilities, and financial side effects are material operating constraints. No maintained first-party general-purpose PayPal CLI or repeatable computer-use result was verified.

Best path today: api · Overall automatability: 9/10