It is opinionated in a useful way: evaluate_change comes first and often calls detect_flag itself, so the common failure — a second flag for something already flagged — is designed out. cleanup_flag closes the other end, which is the half teams usually skip.
A feature-flag workflow over the Unleash Admin API, built for coding assistants: assess a change, check for an existing flag, create one, wrap the code, configure rollout, and remove it later.
- A code change scored for risk, with a recommendation on whether a flag is warranted
- Existing flags discovered before a duplicate is created
- A flag created with validation and typing
- Language-specific guidance on wrapping the change in that flag
- Rollout strategies configured, and a strategy removed from an environment
- A flag enabled or disabled in an environment
- A flag's metadata and activation strategies read back
- Flags in a project and projects available to the token, listed with pagination
- Instructions for safely removing flagged code paths
Node.js 22 or newer and an Unleash instance with an Admin API token — what the server can see and change is bounded by that token's projects.
One command plus a key — npx -y @unleash/mcp@latest --log-level error, then supply credentials
