Connect your agent

ddd-studio exposes a Model Context Protocol (MCP) interface. Any MCP client — Claude Code, Claude Desktop, Cursor, and others — can connect to your account and work on your models: read and edit workspaces, flows, slices, and scenarios, through the same constraint engine the app uses.

Changes made by an agent appear live in your open canvas — no reload. The same goes for a second browser tab. You model on the board, the agent fills in schemas and scenarios, and you review every step visually.

How access works

• The MCP endpoint is https://api.ddd-studio.dev/mcp (Streamable HTTP transport).
• Authentication uses OAuth 2.1 against the same Google sign-in as the web app. On first connect, your MCP client opens a browser window where you sign in and approve access.
• The agent gets the same permissions you have: it sees and edits your workspaces, and nothing else.

Claude Code

Add the server once:

claude mcp add --transport http ddd-studio https://api.ddd-studio.dev/mcp

Then, inside a Claude Code session, run /mcp and complete the sign-in for ddd-studio. From that point on, Claude can list your workspaces and work on them.

Claude Desktop and other clients

In Claude Desktop / claude.ai, add ddd-studio as a custom connector: Settings → Connectors → Add custom connector, with the URL https://api.ddd-studio.dev/mcp. The sign-in flow starts automatically on first use.

Any other MCP client that supports remote servers with OAuth works the same way — point it at the endpoint and complete the browser sign-in.

What to ask an agent to do

The MCP tools are task-oriented, so agents work in the same terms you do. A few prompts that work well:

• "Open my library workspace and summarize the model: contexts, aggregates, invariants."
• "Add a command slice reserve-book to the cataloging context: actor member, command reserve, event reserved."
• "For every invariant in the model, check whether a rejection scenario exists — and add the missing ones with realistic example data."
• "Review the flow order: does any read model receive an event before it happens in the story?"

Because every agent operation passes through the constraint engine, an agent can't corrupt your model any more than you can: invalid references and fields are rejected with clear error messages, and rules like name-uniqueness within a Bounded Context are enforced.

Working in parallel

Agent edits and UI edits are safe to interleave: the studio shows changes from agents (or other tabs) in the open app as they happen. A practical division of labor is to keep the canvas open while the agent works — watch slices and scenarios appear, and intervene visually whenever the model takes a turn you don't like.