Connecting your agent

The config

Every session gets its own address. The demo creates one for you and hands you the finished snippet:

{
  "mcpServers": {
    "mockd-hubspot": {
      "url": "https://mock.mock-d.com/s/<session-token>/mcp"
    }
  }
}

• The section is named the same in every client. Claude, Cursor, the OpenAI Agents SDK, LangGraph — anything that speaks MCP knows it.
• The entry name is yours to pick. Mockd suggests a prefixed one so it stays obvious in your client where the tools came from.
• There's no header and no API key. The token lives in the path.

What happens on connect

Mockd serves the tool side of the protocol: listing tools and calling them. Prompts and resources from the MCP specification aren't exposed. A server that mocks tools doesn't need them.

Arguments are validated

Before a tool call runs, its arguments are checked against the tool's real JSON schema. Send text where a number is expected and your agent gets an error back before anything happens.

That's deliberate. A mock more forgiving than the original lets agents through that would fail in production. This is exactly the class of bug that should surface here.

The connector: an address that stays

A session address covers one run. If you want Mockd to sit in your client config permanently, use a connector instead:

{
  "mcpServers": {
    "mockd-hubspot": {
      "url": "https://mock.mock-d.com/c/<connector-token>/mcp"
    }
  }
}

The address stays constant, the run behind it doesn't. If the last one is still active you join it; if it's over, the next tool call starts a fresh session with fresh data.