Skip to main content

​
Install the Agentcard MCP

One command. A browser sign-in appears the first time a tool runs, and new emails get an account on the spot. If you don’t use Claude Code, it prints the server URL to add in any MCP client. There is one Agentcard MCP server, at https://mcp.agentcard.sh/mcp. What it exposes depends on the credential you connect with: your own account, your organization, or one of your users.

​
Connect as your organization

You can run your whole Agentcard integration without leaving your coding agent. Connect with the same credentials you use for the API:
That works in any MCP client. For Cursor or Claude Desktop, add the same URL and header to your MCP config. Whether you get sandbox or production is decided by the credential you connect with, like everywhere else.

​
What your agent can do with it

22 tools, the same operations as the API. Your agent can build the integration, test it, and debug it in production. When something breaks during your integration, connect this server and ask your agent to debug it. It can read the same state we see.

​
Give your users’ agents a way to buy

The organization server is for you. Your users’ agents connect to a different server, as the user. Create a buy token for a user with mint_buy_token (or POST /api/v1/cardholders/:id/buy_token). It lasts 30 days and only works for that one user. The agent then connects with it:
That agent gets the user’s tools, and the one that matters is buy. The whole story is on the Purchase API page. Next: Test in sandbox