Why use Trio MCP?
- Nothing to build. Add the server to the AI client your team already uses and sign in with your Trio App user. There is no code, no API key to rotate and no server to host.
- Built for agents. Every tool describes when to use it and which tool to prefer for a similar question, amounts come in cents, dates in ISO 8601 and listings are paginated with cursors, so the agent covers the whole period before it answers.
- Scoped to your organization. Every call runs with the credentials that authenticated the session and sees that organization’s data only. Card numbers and CPFs arrive masked.
- Read-only by design. Payments keep going through the approval rules of your account in the Trio App.
What can you do with Trio MCP?
The Supported tools page lists every tool with its parameters and a sample prompt.
How it works
- You add
https://mcp.trio.com.br/mcpto your AI client and sign in to Trio in your browser. Trio issues the client a token limited to your organization. - You ask a question. The client sends it to the model together with the list of Trio tools and their descriptions.
- The model picks a tool, the client calls it on the Trio MCP server with your token, and the server answers with the same data the API returns.
- The model reads the result, calls more tools if it needs to, and writes the answer.
What Trio MCP does not do
- It does not create payments, transfers or refunds, and it cannot approve anything. Money movement stays in the Trio App, under your approval rules.
- It does not issue, block or top up cards, and it does not change account settings or webhooks.
- It does not see other organizations. A token is bound to the organization whose user signed in.
Who can use it
Trio MCP is live for every Trio customer. Any user of your organization in the Trio App can connect an AI client and sign in; the agent then reads the data that user’s organization holds. There is nothing to request or enable first.Next steps
- Connecting Trio MCP: add the server to Claude, Claude Code, ChatGPT, Cursor or any MCP client, and read the security practices before you grant access.
- Supported tools: what each tool returns and the prompts that trigger it.

