Stripe

What it is

Online payments: checkout pages, payment links, invoices, subscriptions and disputes. Many of our booking, deposit and invoicing workflows take money through Stripe so the business never touches card numbers.

How an agent connects

1. Stripe's own MCP server at https://mcp.stripe.com. The owner signs in with OAuth and grants access per account and per environment (live or sandbox) with different permissions for each. An admin can turn MCP access on or off for the whole team in the Dashboard and revoke sessions.
2. Agent API keys for clients that cannot use OAuth: a key with an "Agent" badge and only the permissions the job needs. From 31 October 2026 the MCP server no longer accepts full secret keys or restricted keys without the Agent tag.
3. The API, for apps an intern builds (Checkout, Payment Links, webhooks), always tested in a sandbox first.
4. n8n and Zapier both have Stripe connectors.

What agents may do

agent_access: draft. An agent may read payments, invoices and disputes, and prepare drafts: a draft invoice, a quote, a payment link, a dispute evidence packet. A person finalises and sends invoices, issues refunds and submits dispute evidence.

Through the gateway

The gateway has no Stripe action yet. Until it does, the agent uses the MCP server in a sandbox, or a read-only agent key, or exports. A draft for the live account goes by webhook.post to the owner's automation, always held for approval. Stripe adds its own check: some writes through MCP, such as refunds and outbound payments, need a person to confirm on a Stripe page. Refunds and payouts are never automated.

Watch out for

• Turn on human confirmation for tools in the agent, as Stripe advises, and do not combine the Stripe server with untrusted servers (prompt injection).
• Limits from Stripe's page: 100 requests per second in live mode, 25 in a sandbox, 25 per endpoint by default.
• Webhooks must be verified and deduplicated by event id.
• Disputes allow one evidence submission. A person submits it.