PlannedPlanned. In discovery
Studio in your agent’s loop.
A Studio MCP server, so Claude Code, Cursor or your own agent can ask Studio for status, read a session or send the team a message, under your identity, your persona access and your spend cap.
Planned tools.
Read tools plus one write: sending a message. Descriptions written for an agent reader.
- List your projects and a project's sessions.
- Read a session, paged, from where you left off.
- Start a session with a persona, and send it a message: that starts a turn as you, within your spend cap.
- Get build and deploy status.
- See what needs you, with a link back into Studio.
Rules it will keep.
No decisions over MCP
Anything that needs a person's sign-off links back to Studio. There is no tool that resolves an approval, and a contract test keeps it that way.
On the record
Every call is an audit event marked as coming through MCP, with the token that made it.
Your token, your limits
A personal API token as the bearer; rate limits per token.
Where it starts.
Studio already speaks MCP in both directions: each project connects its own MCP servers (the per-project MCP registry), and Studio runs an MCP server internally for its sandboxed build sessions. The public server reuses that work on top of the planned customer API.
Running an agent stack?
Tell us which client you use. It helps us order the work.