Agents and MCP
How agents will run the same loop a person runs in the console. Coming in a later phase.
Coming in Phases 2, 3 and 7
Agent tokens and approvals arrive in Phase 2. The CLI commands and the MCP server over stdio arrive in Phase 3. The MCP server over HTTP and a Claude Code plugin arrive in Phase 7. Nothing on this page is live yet.
Bandwise is built headless first. Every capability of the console is an operation that the HTTP API, the bandwise CLI and the MCP server also expose, under the same permission checks. An agent can therefore build a set, run evals, publish and move it through rollout.
The plan in brief
- Agent tokens. An agent acts as one user in one org, with scopes and a role ceiling. Its effective role is never higher than the user's own role in that org.
- Agents propose, people approve. High-risk moves by an agent, such as moving a set to
full, wait for a person to approve them. Moves toward safety, such as pausing or rolling back, never wait. - MCP server. A curated set of tools over the same operations, for agents that speak the Model Context Protocol.
This page will cover setup and the tool list when they ship.
Bandwise is an independent product built on TypeSafe's System One models. It is not TypeSafe's documentation. For the System One models themselves, see docs.typesafe.ai.