Connecting a payment tool
How to give Claude or ChatGPT a payment tool via MCP
Add a remote MCP payment server to Claude Desktop, Claude on the web or ChatGPT: which auth each host supports and what to check before it can spend.
Run or connect to an MCP server that exposes payment tools, then add it to the host: as a remote connector with a URL and credential, or as a local stdio server in the desktop app's config. The server, not the chat model, should hold the limits and the money path.
Why is this harder than it sounds?
MCP makes adding a tool easy, which is the risk. Once a payment tool is connected, the model decides when to call it, based on whatever is in the conversation, including documents and web pages. Hosts also differ in how they authenticate to a remote server: some send a fixed header, some only do OAuth, some need a local bridge process. Picking the wrong route either fails silently or leaves a credential somewhere you didn't intend. And the host's own 'allow this tool' prompt is a generic consent screen, not a check on amount or recipient.
How do you do it?
- Pick a server that enforces policy itself. The MCP server should apply limits and approvals server-side and expose only narrow tools such as create_payment and get_budget. Avoid servers that ask you to paste a private key into their config.
- Claude Desktop: use a local stdio entry. Add the server to claude_desktop_config.json under mcpServers with a command, arguments and environment variables. A bridge process can hold the credential locally and relay calls to a hosted server.
- Claude on the web: add a custom connector. Claude's help centre says custom connectors take a remote MCP URL and support OAuth, no sign-in, or fixed request headers such as a bearer token. Free plans are limited to one custom connector, and on Team and Enterprise everyone using the connector shares its credential.
- ChatGPT: check what its connector form can send. ChatGPT's support for custom MCP connectors and their authentication options has changed several times. Before you rely on it, check OpenAI's current documentation for whether it can send a fixed Authorization header or needs OAuth.
- Start on a simulated source. Connect with a source that moves no money and ask the assistant to do something that costs a little. Watch the tool calls and the policy decisions.
- Keep host confirmations on. Leave the host's per-call confirmation enabled for the payment tool, at least until you trust the setup. It is a second, coarse check on top of the server's policy.
What it does not mean
- Connecting a payment tool to ChatGPT is not the same as ChatGPT's own shopping checkout. Merchant checkout in ChatGPT runs on the Agentic Commerce Protocol, which is a different integration for sellers.
- The host's tool permission prompt is not a spending limit. It cannot tell a $2 payment from a $2,000 one.
- A credential in a shared team connector is shared. Everyone who uses it spends under the same policy.
Common questions
Can I use the same Veyra endpoint in several hosts?
Yes, the credential works anywhere that can send it. Every host then spends under the same limits and daily cap, so consider one endpoint per agent instead.
Does the chat model ever see my wallet?
Not with Veyra. list_payment_sources returns masked sources only, and no tool returns a key or seed phrase.
Is this financial advice?
No. It explains how to connect tools and keep them under control, not what to spend money on.
Where does Veyra fit?
Veyra's endpoint is https://veyra.money/api/mcp and authenticates with a static header, Authorization: Bearer veyra_…; it does not offer OAuth sign-in. For Claude Desktop, Veyra's docs use the open-source veyra-mcp stdio bridge (npx -y veyra-mcp with VEYRA_TOKEN in env), which relays every call to the hosted endpoint and holds no policy itself. Claude's custom connectors can send fixed request headers, which is what the endpoint expects, but Veyra's docs don't yet walk through that route; if it gives you trouble, use the Desktop bridge. Veyra has not published a ChatGPT setup.
Sources
Terms on this page
- MCP serverAn MCP server is a service that offers tools, resources or prompts to AI applications using the Model Context Protocol, an open protocol built on JSON-RPC 2.0.
- Shared Payment Token (SPT)A Shared Payment Token (SPT) is a Stripe credential that grants an agent or seller scoped use of a customer's payment method, such as a card, with usage and expiration limits.
This page explains how agent payments and the related technology work. It is not financial, investment, legal or tax advice. Protocols and provider features change; check the linked official sources before relying on a detail.