Governed agent operations through MCP.
AI coding agents can execute quickly but still fail quietly. MartinLoop MCP adds a governed workflow layer so runs are scoped, verified, and easier to review.
Current package line: @martinloop/mcp@0.9.0.
What this solves: agent work that is hard to audit, hard to verify, and easy to overrun. MartinLoop MCP keeps runs grounded in preflight, verifier evidence, and persisted receipts.
Install
npx -y @martinloop/mcp
Host setup
# Codex codex mcp add martin-loop -- npx -y @martinloop/mcp
# Claude Code (macOS / Linux) claude mcp add --transport stdio --scope user martin-loop -- npx -y @martinloop/mcp
# Claude Code (Windows PowerShell / cmd) claude mcp add --transport stdio --scope user martin-loop -- cmd /c npx -y @martinloop/mcp
What you get
- Readiness and preflight checks before spend
- Governed run execution surface with verifier-aware outcomes
- Persisted run inspection and triage views
- Dossier-style review for latest or selected runs
- Host profile guidance for minimal, diagnostic, and full-local operation modes
Operator prompts
- “Preflight this task with a strict verifier and budget cap.”
- “Run a governed attempt and then show dossier-level review output.”
- “Inspect latest verifier evidence and summarize next safe action.”
Same governance, host-native workflow.
Use MCP for host-native operations, then pivot to CLI when you need explicit local run control.