# Refactron

> Prove that any change, your AI agent's, a codemod's, or your own, preserved behavior. Refactron runs your real tests in an isolated shadow tree and lands only on green, with a reproducible record of why it's safe.

## Docs

- [Overview](https://docs.refactron.dev/overview.md): Prove that any change, your AI agent's, a codemod's, or your own, preserved behavior. Refactron runs your real tests in an isolated shadow tree and lands only on green, with a reproducible record of why it's safe.
- [Quickstart](https://docs.refactron.dev/quickstart.md): Install Refactron from npm and verify your first diff end to end, then give the same gate to your AI agent over MCP.
- [FAQ](https://docs.refactron.dev/faq.md): Frequently-asked questions about Refactron: the verification layer for AI code change.
- [Verify a diff](https://docs.refactron.dev/verification/verify-diff.md): Prove a change preserved behavior. Refactron runs your real tests in an isolated shadow tree and returns SAFE, UNSAFE, or UNPROVEN.
- [MCP server](https://docs.refactron.dev/verification/mcp-server.md): The Refactron MCP server has its own tab: setup for every major AI coding client, plus the verify_change tool contract.
- [Verdicts](https://docs.refactron.dev/verification/verdicts.md): SAFE, UNSAFE, and UNPROVEN: the three-way verdict at the center of Refactron, how coverage fuses with the gates, and why UNPROVEN is the honest answer.
- [Safety Model](https://docs.refactron.dev/concepts/safety-model.md): Three gates and a coverage check, run against your change in an isolated shadow tree. The engine that turns a diff into a SAFE, UNSAFE, or UNPROVEN verdict.
- [Deterministic by design](https://docs.refactron.dev/concepts/why-no-llm.md): Generating a change is probabilistic. Deciding whether it's safe has to be deterministic. Refactron is the gate an LLM agent calls to prove its own work.
- [Changelog](https://docs.refactron.dev/changelog.md): Release history for Refactron, newest first.
- [MCP server](https://docs.refactron.dev/mcp/overview.md): Refactron over the Model Context Protocol: one tool, verify_change, that returns SAFE, UNSAFE, or UNPROVEN on a proposed change before an agent lands it.
- [MCP quickstart](https://docs.refactron.dev/mcp/quickstart.md): Install the Refactron MCP server, register it with one client, confirm the tool is listed, and get a first real verdict on a real change.
- [Claude Code](https://docs.refactron.dev/mcp/claude-code.md): Register the Refactron MCP server with Claude Code using the claude mcp add command or a project .mcp.json.
- [Claude Desktop](https://docs.refactron.dev/mcp/claude-desktop.md): Add the Refactron MCP server to Claude Desktop by editing claude_desktop_config.json, then restart the app.
- [Codex CLI](https://docs.refactron.dev/mcp/codex.md): Add the Refactron MCP server to OpenAI Codex CLI. Note the format: Codex reads TOML, not JSON.
- [Cursor](https://docs.refactron.dev/mcp/cursor.md): Add the Refactron MCP server to Cursor with a project .cursor/mcp.json or the global ~/.cursor/mcp.json.
- [VS Code](https://docs.refactron.dev/mcp/vscode.md): Add the Refactron MCP server to VS Code with GitHub Copilot. Note the root key: VS Code uses servers, not mcpServers.
- [Gemini CLI](https://docs.refactron.dev/mcp/gemini-cli.md): Add the Refactron MCP server to Gemini CLI with the gemini mcp add command or ~/.gemini/settings.json.
- [Windsurf](https://docs.refactron.dev/mcp/windsurf.md): Add the Refactron MCP server to Windsurf Cascade by editing ~/.codeium/windsurf/mcp_config.json.
- [Any other client](https://docs.refactron.dev/mcp/other-clients.md): Connect Refactron to any MCP client that can launch a stdio server, plus the root keys and config shapes each client family expects.
- [verify_change reference](https://docs.refactron.dev/mcp/tool-reference.md): The verify_change MCP tool contract: arguments, request and response examples, the error shape, and how it maps to verify-diff exit codes.

## OpenAPI Specs

- [openapi](https://docs.refactron.dev/api-reference/openapi.json)