Basics

Core concepts

Six terms show up everywhere in Mockd. Once they land, the rest of these pages mostly explains itself.

The whole picture in one diagram

Your agent speaks MCP, exactly as it would to a real server. What answers behind the endpoint is your choice.

Mockd isn't an SDK or a library. It's an address you put in your MCP client. Your agent code stays as it is.

The terms

Mock server: An endpoint that mirrors a real MCP server's tools: same names, same descriptions, same JSON schemas. It answers with synthetic data.
Session: A sandboxed run with its own data and its own trace. Created without an account; the token in the URL is the only credential.
Connector: An address that stays constant. Put it in your client once, but every new run still gets a fresh session.
Tool call: One call from your agent to a tool. That's the unit Mockd sees and records.
Trace: The list of every tool call in a session, in the order they happened, with arguments, response, latency and status.
Scenario: A switch that makes every following tool call fail: rate limit, timeout, empty result, or extra latency.

What Mockd sees — and what it doesn't

Mockd sits on the MCP connection. It sees every tool call: which tool your agent picked, with which arguments, in which order, and what came back.

It does not see what your model was thinking. Prompts, system messages and model responses travel between your client and your model provider, not over this connection.