# agentbelt

> agentbelt is an open-source MCP server for developer tools. Shared, audited git, test and workflow tools for MCP coding apps that never push. It runs locally on your machine and is started by your AI client.

- **Kind:** MCP server
- **Pricing:** Open source
- **Platforms:** CLI, macOS, Linux
- **Website:** https://lovark.github.io/agentbelt/
- **Repository:** https://github.com/lovark/agentbelt
- **MCP Registry name:** io.github.LovaRK/agentbelt
- **Categories:** developer tools, security
- **Listing updated:** 2026-11-14

## About

agentbelt is a local stdio server that gives several coding assistants the same toolbelt, so the safety rules are written once instead of per app. It brings no model of its own and needs no API key. Setup runs through npx, installs a pinned copy with install scripts disabled, checks it with real tool calls, then shows and backs up each config change before it edits Claude Code, Codex, Cursor, VS Code, LM Studio and other apps.

The git tools read status, diffs and logs, and start task branches named agentbelt/<id> where commits are made. Push, reset, rebase and clean are blocked. Testing tools detect and run a project's own typecheck, lint, build and tests, and workflow tools run recipes such as check, repo-health and safe-commit. Approval steps can only be approved by you with agentbelt approve in a terminal, and every call is written to .agentbelt/audit.jsonl with secrets removed.

The site labels built-in tools as policy-only, meaning checked but not OS-sandboxed, while third-party plugins must run in a sandbox with no network. It needs Node.js 22.13 or newer and git on macOS or Linux, and nothing updates automatically.

## Features

- **Isolated task branches:** Commits go only to agentbelt/<id> branches, and pushes are blocked.
- **Project checks:** Detects and runs the repository's own typecheck, lint, build and tests.
- **Approval-gated workflows:** Recipes such as repo-health pause for approval that only the user can give.
- **Audit log:** Every call is recorded to a local JSONL file with secrets removed.
- **Safe setup:** Backs up each app config and verifies a pinned copy before switching.
- **Update and rollback:** Updates install beside the old version and can be rolled back instantly.

## Good for

- Let an agent run lint and tests and fix failures without push access
- Review uncommitted changes with the same tools in any app
- Keep one audit trail across several AI coding apps
- Run a safe-commit workflow that needs your approval

## Install

### Claude Code (Terminal)

```sh
claude mcp add agentbelt -- npx -y agentbelt
```

### Claude Desktop (claude_desktop_config.json)

```json
{
  "mcpServers": {
    "agentbelt": {
      "command": "npx",
      "args": [
        "-y",
        "agentbelt"
      ]
    }
  }
}
```

### Cursor (~/.cursor/mcp.json)

```json
{
  "mcpServers": {
    "agentbelt": {
      "command": "npx",
      "args": [
        "-y",
        "agentbelt"
      ]
    }
  }
}
```

### VS Code (.vscode/mcp.json)

```json
{
  "servers": {
    "agentbelt": {
      "type": "stdio",
      "command": "npx",
      "args": [
        "-y",
        "agentbelt"
      ]
    }
  }
}
```

### Windsurf (~/.codeium/windsurf/mcp_config.json)

```json
{
  "mcpServers": {
    "agentbelt": {
      "command": "npx",
      "args": [
        "-y",
        "agentbelt"
      ]
    }
  }
}
```

### Other clients (your client's MCP settings (most use this shape))

```json
{
  "mcpServers": {
    "agentbelt": {
      "command": "npx",
      "args": [
        "-y",
        "agentbelt"
      ]
    }
  }
}
```

## Tools

- `git_status`: Reads repository status.
- `git_diff`: Reads repository diffs.
- `git_log`: Reads repository history.
- `git_task_start`: Starts an isolated task branch.
- `git_commit`: Commits to a task branch; never pushes.
- `testing_detect`: Detects the project's checks.
- `testing_run_checks`: Runs the project's typecheck, lint, build and tests.
- `workflows_list`: Lists available workflows.
- `workflows_start`: Starts a workflow such as repo-health.
- `workflows_show`: Shows a workflow.
- `agentbelt_status`: Reports the workspace, plugins and command isolation.

## FAQ

### What does it need?

Node.js 22.13 or newer and git, on macOS or Linux.

### Can the agent push code?

No. Push, reset, rebase and clean are blocked, and commits go only to task branches.

### Which apps work?

Claude Code, Codex, Cursor, VS Code, DeepSeek Harness and LM Studio were tested with real calls; Claude Desktop and Goose are configured but not yet verified.

### Does it need an API key?

No. Your AI app supplies the model.

### Is agentbelt free?

Yes. agentbelt is open source, so you can use and self-host it for free.

### How do I add agentbelt to Claude Code?

Run this command in your terminal: claude mcp add agentbelt -- npx -y agentbelt

### What are the best alternatives to agentbelt?

The closest alternatives to agentbelt are REA, Codebase Memory, GitHub, Serena MCP: the IDE for your agent and XERJ.

## Alternatives

- [REA](https://ai-powered.com/mcp/rea): REA is an open-source MCP server for developer tools.
- [Codebase Memory](https://ai-powered.com/mcp/codebase-memory): Codebase Memory is an open-source MCP server for knowledge and memory.
- [GitHub](https://ai-powered.com/mcp/github-2): GitHub is an open-source MCP server for developer tools.
- [Serena MCP: the IDE for your agent](https://ai-powered.com/mcp/serena-mcp-the-ide-for-your-agent): Serena MCP: the IDE for your agent is an open-source MCP server for developer tools.
- [XERJ](https://ai-powered.com/mcp/xerj): XERJ is an open-source MCP server for knowledge and memory.
- [Ouroboros](https://ai-powered.com/mcp/ouroboros): Ouroboros is an open-source MCP server for developer tools.
- [Claude Code Ultimate Guide](https://ai-powered.com/mcp/claude-code-ultimate-guide): Claude Code Ultimate Guide is an open-source MCP server for ai models and llm tooling.
- [Concord MCP](https://ai-powered.com/mcp/concord-mcp): Concord MCP is a freemium MCP server for developer tools.

---
Source: https://ai-powered.com/mcp/agentbelt
