generated: '2026-09-19'
method: searched
source: https://pypi.org/project/rosentic-mcp/
docs:
- https://rosentic.com/mcp/
- https://rosentic.com/docs/
clis:
- name: rosentic-mcp
  purpose: >-
    The rosentic-mcp binary is both the local MCP server (run with no arguments, stdio) and Rosentic's
    command-line surface for enforcement and diagnostics: install committed git and Claude Code hooks, run
    the merge gate directly in CI or by hand, and check which server and engine are answering.
  language: Python
  package: https://pypi.org/project/rosentic-mcp/
  version: 0.2.9
  repository: https://github.com/Rosentic/rosentic-action
  docs: https://pypi.org/project/rosentic-mcp/
  install:
  - method: pip
    steps: [pip3 install --upgrade rosentic-mcp]
    note: '"Use --upgrade even on a machine that already has the package: a plain pip install is a no-op for returning users and silently keeps stale code."'
  - method: uvx
    steps: ['uvx --from rosentic-mcp rosentic-mcp']
  prerequisites: [Python 3.11+, git, 'Docker Desktop running (pulls ghcr.io/rosentic/rosentic-engine on first scan) or ROSENTIC_ENGINE_PATH pointing at a local engine checkout']
  invocation: rosentic-mcp [command] [flags]
  commands:
  - group: server
    commands:
    - name: (no arguments)
      description: Start the MCP server on stdio — what {"command":"rosentic-mcp"} in an MCP client config launches.
  - group: enforcement
    commands:
    - name: install-hooks
      description: >-
        Write a committed .githooks/pre-push that runs the gate and exits nonzero on UNSAFE, set
        git config core.hooksPath .githooks for this clone, and add Claude Code hooks to .claude/settings.json
        (PreToolUse blocks an UNSAFE write; Stop keeps the agent working while UNSAFE findings exist).
        Idempotent; appends to an existing settings.json and never removes the user's entries.
      usage: 'rosentic-mcp install-hooks && git add .githooks .claude/settings.json'
    - name: gate
      description: Run the merge gate directly. Exit 0 clean, 1 UNSAFE, 2 scan error. Fails closed — if the scan cannot run the push is blocked with an explanation.
      usage: 'rosentic-mcp gate --repo . --base main'
      flags: [--repo, --base]
      env: ['ROSENTIC_SKIP=1 bypasses one push', 'ROSENTIC_STRICT=1 makes WARNING findings block too']
  - group: diagnostics
    commands:
    - name: doctor
      description: Report which server answered, which engine it resolved, and whether that combination is supported. Names a source checkout answering instead of the installed package.
      usage: 'rosentic-mcp doctor   # --json for machine-readable; exits 1 when unsupported'
      flags: [--json]
  environment:
  - {name: ROSENTIC_API_KEY, purpose: 'workspace key (ros_live_ prefix) for get_verdict / get_remediation_queue and dashboard history', required: false}
  - {name: ROSENTIC_ENGINE_PATH, purpose: 'local engine checkout instead of the Docker image', required: false}
  - {name: ROSENTIC_SKIP, purpose: 'bypass one gated push', required: false}
  - {name: ROSENTIC_STRICT, purpose: 'make WARNING block at the gate', required: false}
  key_flows:
  - name: Commit the enforcement gate to a repo
    steps: [pip3 install --upgrade rosentic-mcp, rosentic-mcp install-hooks, 'git add .githooks .claude/settings.json && git commit -m "Add Rosentic enforcement gate"']
  - name: Verify the install answers from the package
    steps: [rosentic-mcp doctor, 'claude mcp get rosentic']
related:
  github_action: 'uses: Rosentic/rosentic-action@v1  (the CI backstop; inputs in packages/rosentic-com-packages.yml)'
  engine_cli: >-
    The docs' orchestrator page also shows "python3 detect.py check-siblings <repo> --base main --branches a,b,c
    --format json", the engine's own entrypoint inside the container; it is not distributed as a standalone
    binary and is recorded here only because the provider documents it as the Tier-3 "CLI available now".