Skip to main content
If you build with an AI assistant, you can connect it to PontisGlobe directly and integrate by describing what you want in natural language. Aurora answers questions about what we support, looks up what a payout to a given country needs, and checks a request you have built against the same validator our API runs — before you send anything. Free, and it connects in under a minute.

​
Two servers

Add both. The first checks what you build; the second explains how anything works.

​
Add it

Claude Desktop and ChatGPT take the URL through their own UI rather than a file — look for Connectors or Settings → MCP, choose a custom or remote server, and paste https://mcp.pontisglobe.com/mcp.
Config keys differ between clients and move between versions — servers in VS Code, serverUrl in Windsurf, httpUrl in Gemini CLI. If one of these does not take, your client’s own MCP docs are the authority. Nothing on our side changes: it is the same URL every time.

​
Check it worked

Most clients list the servers they are connected to. In Claude Code or Codex CLI, run /mcp — you should see PontisGlobe Aurora. Then just ask a question.

​
Sample prompts

Ask in plain language. Start here:
What does PontisGlobe do? Which countries can I pay out to? What does it cost?
Then, once you are building:
What fields does a mobile money payout need? Build a payout for 250 USDT to a Nigerian bank account and check it. Why is this request being rejected?
Or hand over the whole job:
Add PontisGlobe payouts to my Node app — signing, the request, and the callback handler. Set up a checkout that collects USDT and tells my server when the customer has paid. Port my payout integration from Python to Go.
With both servers connected your assistant has the reference and the validator at once, so it can write the code, check its own request against the real schema, and fix it before you run anything.
If your assistant answers from memory instead of checking, say so explicitly: “Using the PontisGlobe MCP server, …”. Once it has called a tool a couple of times it tends to keep doing it.
Your assistant picks the right tool on its own, gets a real answer rather than a guess, and corrects its own mistakes before you send a single request. If you want to see everything on offer, ask it to list the tools — we add to them as the product grows. One question is worth asking early: the fields a recipient needs differ per payment method, and getting them wrong is the most common reason a first payout is rejected. Ask what a method needs before you build the request, not after it comes back rejected.

​
Is it safe to connect?

Yes — everything here is public and account-free. There is no key to hand over, and nothing behind it that belongs to you: it cannot read a balance, look up a payment, list your beneficiaries or move money. It answers questions about the product and about the shape of a request. Account-aware tools are coming. Those will ask you to connect your account first, and until you do, this stays exactly as safe as it is today.
The validators behind these tools are the same ones our API runs, so a request that passes here will not be rejected for its shape. It can still be refused for a business reason — insufficient funds, a corridor not enabled on your account — which needs a real call. See Errors.