Use paperplane in Microsoft Copilot Studio

Copilot Studio added native Model Context Protocol support with an onboarding wizard that discovers a server’s tools automatically — no manual action-by-action setup, and no OpenAPI file needed for a plain, unauthenticated remote server like ours. To be clear about what this is: this is user-side setup. paperplane has no native listing in Copilot Studio’s tool catalog today — you are pointing your agent at our standard MCP endpoint yourself, the same endpoint Claude, ChatGPT, Grok, and every other MCP client uses.

Connect it (MCP onboarding wizard)

• Open your agent in Copilot Studio and go to the Tools page.
• Select Add a tool, then New tool, then Model Context Protocol.
• Fill in a Server name (e.g. “paperplane”), a short Server description (the orchestrator uses this to decide when to call it — something like “quotes, sends, and tracks real USPS letters” works), and the Server URL: https://sendpaperplane.com/api/mcp.
• For authentication, choose None — our server has no API key or OAuth requirement — and select Create.
• On the Add tool dialog, select Create a new connection, then Add to agent. Copilot Studio discovers all three tools: quote_letter, send_letter, and get_letter_status.

What the three tools do

• quote_letter — read-only and free. Prices a letter (First-Class $1.99, Certified $12.99, Certified + electronic Return Receipt $14.99, Priority $24.99) or a piece format (postcards from $1.99, the folded notecard $5.99), and — when given the real recipient and content — mints a single-use confirmation_token bound to that exact piece.
• send_letter — the one write tool: it spends money and mails a physical object. It refuses without a valid confirmation_token from a matching quote, so the agent can never mail something as its first move. Live sends return a Stripe payment link for a human to approve; sandbox: true completes instantly, free, mailing nothing.
• get_letter_status — read-only tracking: screening → printing → mailed → delivered.

Sandbox first: a safe walkthrough

Once the tool is added, try a prompt like this in the agent’s test pane:

"Send a one-page test letter to Maria Alvarez, 1 Main St, Richmond VA 23220
in sandbox mode. Show me the price before you send anything."

The agent should call quote_letter first and show you a price with a confirmation_token, then call send_letter with that token and sandbox: true. The sandbox order runs the entire real pipeline — address checks, content screening, simulated printing and tracking — but nothing is printed, mailed, or charged. When you are ready for a real letter, drop the word “sandbox” and approve the Stripe payment link the agent hands back.

If you would rather use a custom connector

Copilot Studio’s second path — a custom connector built in Power Apps from an OpenAPI schema — exists mainly for MCP servers that need it (data-loss-prevention policy scoping, for example). paperplane’s MCP server is a plain streamable-HTTP endpoint with no auth, so the onboarding wizard above is simpler and is what we recommend. If your organization requires routing MCP access through a governed custom connector anyway, our full REST API (POST /v1/quotes, POST /v1/orders, GET /v1/orders/:id) is documented and stable enough to build a connector schema against directly.

If you would rather not use MCP

The same backend is a plain REST API, documented on the developer docs. For the full mechanics of the quote→send confirmation contract, see how AI agents mail letters via MCP, or watch it happen live in the no-signup demo.