Martin Loop
Run coding agents with budgets, verifier checks, recovery and a clear handoff
- Platforms
- CLI
- Runs
- Locally
- Version
- 0.9.0
- Listing updated
- today
Similar MCP servers
About Martin Loop
Martin Loop is an MCP server for developer tools. It runs locally on your machine and is started by your AI client.
MartinLoop wraps the coding job rather than writing code itself. You set an objective, verifier command, spend ceiling (--budget-usd) and iteration cap, and the job ends with VERIFIED, STOPPED or NEEDS REVIEW plus receipts. The README stresses that VERIFIED only proves the checks that were configured.
Its audit command reads Claude Code session history locally to report API-equivalent spend inside fix-and-retry loops, with an offline option. Work is local-first, with optional hosted dashboard sync. It supports runtimes such as Codex, Claude and Gemini, and passes any --model through unchanged. The @martinloop/mcp package adds the MCP interface.
Install
Run in a terminal
claude mcp add martin-loop -- npx -y @martinloop/mcpAdd to claude_desktop_config.json
{
"mcpServers": {
"martin-loop": {
"command": "npx",
"args": [
"-y",
"@martinloop/mcp"
]
}
}
}Add to ~/.cursor/mcp.json
{
"mcpServers": {
"martin-loop": {
"command": "npx",
"args": [
"-y",
"@martinloop/mcp"
]
}
}
}Add to .vscode/mcp.json
{
"servers": {
"martin-loop": {
"type": "stdio",
"command": "npx",
"args": [
"-y",
"@martinloop/mcp"
]
}
}
}Add to ~/.codeium/windsurf/mcp_config.json
{
"mcpServers": {
"martin-loop": {
"command": "npx",
"args": [
"-y",
"@martinloop/mcp"
]
}
}
}Add to your client's MCP settings (most use this shape)
{
"mcpServers": {
"martin-loop": {
"command": "npx",
"args": [
"-y",
"@martinloop/mcp"
]
}
}
}Using several servers? Build one config file with the MCP config generator.
What it does
- Budget and iteration limits
- Hard spend and attempt ceilings per run.
- Verifier evidence
- Completion needs fresh checks bound to the run.
- Recovery state
- Keeps rollback evidence for retries or review.
- Loop-tax audit
- Measures spend in fix-and-retry loops.
- Dossiers and history
- Inspect results, failure classes and stored integrity.
Good for
- Let a coding agent run a bounded job unattended
- Measure retry spend from Claude Code sessions
- Keep scope limits on agent edits
- Hand off a run with receipts
Martin Loop FAQ
Does it write code?
No; the coding agent does.
Which runtimes?
Codex, Claude and Gemini are named.
Is data sent anywhere?
Local-first; hosted sync is optional.
How do I start?
Run `npx -y martin-loop@latest start`.
How do I add Martin Loop to Claude Code?
Run this command in your terminal: claude mcp add martin-loop -- npx -y @martinloop/mcp
What are the best alternatives to Martin Loop?
The closest alternatives to Martin Loop are REA, Context7, Chrome DevTools MCP, Codebase Memory and GitHub.
Where this comes from
- Description: written from the project's README.
- Install steps: from the official MCP Registry entry.


