@martinloop/mcp
Advanced tools
Governed MCP server for AI coding agents with budgets, verifier gates, and inspectable runs.
Stop runaway loops, bad code, and token waste.
The governed MCP runtime for Claude Code, Codex, Gemini CLI, Cursor, and autonomous coding agents.
AI coding agents can write code. MartinLoop decides whether they are allowed to keep spending, whether the result is actually verified, and what evidence must remain for review.
It wraps agent execution with:
| Host | Command |
|---|---|
| Claude Code | claude mcp add martin-loop -- npx -y @martinloop/mcp |
| Codex | codex mcp add martin-loop -- npx -y @martinloop/mcp |
| Gemini CLI | gemini mcp add martin-loop -- npx -y @martinloop/mcp |
| Any stdio MCP host | npx -y @martinloop/mcp |
Windows Claude Code:
claude mcp add --transport stdio --scope user martin-loop -- cmd /c npx -y @martinloop/mcp
No API key is required to start and inspect the server. Live execution requires a supported coding-agent CLI and its normal authentication.
| Uncontrolled agent run | Governed MartinLoop run |
|---|---|
| The agent decides when it is done | Your verifier decides whether completion is valid |
| Retries continue until the user notices | Hard budgets and stop conditions prevent open-ended spend |
| Scope drifts across the repository | Allowed and denied paths constrain writes |
| The final answer hides failed attempts | The dossier preserves attempts, costs, verification, and artifacts |
| Process control is ephemeral | Pause, continue, and cancel actions leave durable receipts |
| “Probably fixed” becomes accepted | Missing, failed, contradicted, or unknown verification stays incomplete |
MartinLoop is not another coding agent. It is the enforcement and evidence layer around the agents you already use.
Use this sequence for governed coding work:
martin_doctor
→ martin_estimate
→ martin_plan
→ martin_preflight
→ martin_run
→ martin_dossier
→ martin_eval
martin_doctor to inspect the environment and run store.martin_estimate before spending.martin://agent/next-step or martin://guides/agent-start when the next action is unclear.martin_create_pr with execute: true unless the user has explicitly authorized a GitHub write.martin_run intentionally hard-blocks until the required doctor, estimate, plan, and preflight receipts exist for the same task.
User objective:
"Fix the auth regression. Budget: $3. Verify: npm test."
doctor checks environment, CLI availability, and run storage
estimate predicts route, spend, and pre-work burn without spending
plan defines scope, verifier proposal, policy pack, and risk
preflight validates the contract before execution
run executes inside budget, scope, policy, and verifier gates
dossier returns attempts, cost, verification, artifacts, and evidence paths
eval grades completion, verifier health, risk, and reviewability
Every attempt is bounded by the run contract. A confident model answer does not override a failed verifier.
This real governed run spent $0.51 against a $3.00 budget. The verifier passed and the receipt integrity was signed. MartinLoop still kept the result at EVIDENCE_BOUNDARY because rollback evidence had not been recorded.
Inspect the example receipt:
martin_doctor — inspect environment readiness; expected first callmartin_estimate — estimate cost and route without spendingmartin_plan — produce scoped plan, verifier proposal, policy pack, and riskmartin_preflight — validate readiness before executionmartin_run — execute a governed coding taskmartin_pause — record a durable pause requestmartin_continue — record a durable resume requestmartin_cancel — record a durable cancellation requestmartin_status, martin_logs, martin_inspectmartin_list_runs, martin_triage_runsmartin_get_run, martin_get_attempt, martin_get_verification_resultsmartin_run_dossier, martin_dossier, martin_evalmartin_pr_summary, martin_create_pr, martin_review_prThe server exposes run summaries, receipts, budget status, verifier evidence, policies, health, agent guidance, and repository-risk context through martin:// resources, including:
martin://runs/latest
martin://runs/latest/summary
martin://runs/latest/receipt
martin://runs/latest/budget-status
martin://runs/latest/verifier-evidence
martin://runs/recent
martin://server/health
martin://policies/current
martin://agent/next-step
martin://guides/agent-start
martin://repo/risk-map
Use resources when the agent needs context without side effects.
Generate host config tuned to the minimum surface needed:
npx martin-loop mcp print-config --host claude --profile minimal
npx martin-loop mcp print-config --host claude --profile diagnostic
npx martin-loop mcp print-config --host claude --profile full-local
npx martin-loop mcp print-config --host claude --profile github-review
minimal — core run and inspection pathdiagnostic — adds doctor and triage surfacesfull-local — all local toolsgithub-review — adds PR evidence and review workflowApache-2.0.
FAQs
Governed MCP server for AI coding agents with budgets, verifier gates, and inspectable runs.
The npm package @martinloop/mcp receives a total of 91 weekly downloads. As such, @martinloop/mcp popularity was classified as not popular.
We found that @martinloop/mcp demonstrated a healthy version release cadence and project activity because the last version was released less than a year ago. It has 1 open source maintainer collaborating on the project.
Did you know?
Socket for GitHub automatically highlights issues in each pull request and monitors the health of all your open source dependencies. Discover the contents of your packages and block harmful activity before you install or update your dependencies.
Company News
Open source maintainers are under more pressure than ever. We're raising our open source program from the Team plan to the Business plan, free.
Security News
The supply chain control that delays freshly published gems now covers lockfile generation and gem vendoring in Ruby projects.
Security News
During a UK cyber test, a Mythos 5 agent used sockpuppets, social engineering, and prompt injection to try to get a maintainer to merge malware.