Skip to main content
SudoMock has a remote MCP server at https://mcp.sudomock.com. A client that speaks MCP connects to it once, you sign in through your browser, and the agent can then list your templates and render them from inside the conversation. The same account and the same credits sit behind it, so anything you render this way is billed exactly as an API render is.
Sign-in happens in the browser. No client asks you to paste a password, a one time code or an API key into the conversation, and no client needs your sm_ key to use the server.

​
Add the server

In Claude Code, open /mcp afterwards and complete authentication if it is requested. Some clients only publish new tools after a reconnect or a restart. For any other client, use its own remote HTTP MCP configuration with OAuth and point it at the same URL. If a client cannot be configured from inside a session, add the server from its settings instead.

​
Install the skill

SudoMock publishes a skill file at https://sudomock.com/docs/skill.md. It tells an agent what this product can do, what a call needs, and what an account allows, so the agent gets it right on the first try.

Install the SudoMock skill into your agent.

Open in Cursor
A client connected to the MCP server above already reads the same file as a resource. This step is for the clients that are not.

​
Let an agent do it

Hand the agent this file and it will work out the client it is running in, add the server, and stop at the sign-in step:
It covers the read-only check below as well, so a setup run generates nothing and spends nothing.

​
Check the connection

Call get_account. An authenticated response means the connection is live. Then confirm the tools for your own work are present: Discovery and get_account are enough. Nothing else needs to run to prove the connection, and a render, an upload or a webhook call would spend credits to tell you something you already know.

​
What the server publishes

The server publishes tools only. It exposes no resources and no prompts, so a client that asks for them gets nothing back. Some tools answer to a second, older name. Use whichever the connection offers and read the schema it gives you rather than assuming it mirrors the REST endpoint of the same name. There is no batch render tool: one call produces one output, and several outputs mean several calls.

​
Before you promise a result

An account on trial credits renders at most 1024 px wide and its output carries a watermark. Check the account with get_account before you promise a size.

Use an API key instead

Write the calls yourself from a backend or a job runner.

Errors

Every error code, and which statuses are worth retrying.

Usage limits

The request rate, the parallel ceiling, and the headers.