| FMRS | 51 / 100 · D | 79 / 100 · B |
|---|
| Reliability | 8 / 20 | 12 / 20 |
|---|
| Security and permissions | 11 / 20 | 18 / 20 |
|---|
| Maintenance | 10 / 20 | 16 / 20 |
|---|
| Documentation | 13 / 20 | 18 / 20 |
|---|
| Setup experience | 9 / 20 | 15 / 20 |
|---|
| Best for | - Developers who want ChatGPT to actually edit and execute against a local codebase
- Advanced users comfortable configuring a tunnel, a browser extension and per-capability permission switches in a beta product
- Users on Windows 10/11, macOS 13 Ventura or newer, or a current desktop Linux with Chrome 116+ or Edge
| - Engineering teams scaling AI coding agents while controlling spend
- Platform teams that need auditable, rollback-aware agent execution records
- Organizations using multiple agents (Claude, Codex, Gemini) under one governed flow
|
|---|
| Not for | - Users who want a zero-config, officially hosted remote MCP service
- ChatGPT workspaces that cannot enable Developer mode and custom MCP apps
- Anyone needing a true OS-level sandbox: commands run with your logged-in user privileges and desktop control is not folder-scoped
- Linux environments without a Secret Service keyring such as GNOME Keyring or KWallet
| - Developers who just want a simple code-completion helper without run governance
- Teams unwilling to accept local JSONL run records with HMAC signing
- Environments without a supported adapter (Claude/Codex/Gemini CLIs, direct provider, or verifier-only)
|
|---|
| Required permissions | - File read/write limited to folders you approve in Settings → Workspace, with paths validated and canonicalised first
- Command execution as your logged-in user, not folder-sandboxed
- Desktop control (on by default on Windows, opt-in on macOS with Screen Recording and Accessibility grants) applying to the whole desktop
- Browser extension running only on chatgpt.com, chat.openai.com and the app's loopback bridge, recording conversation state locally
- OS secret storage: DPAPI on Windows, Keychain on macOS, libsecret or KWallet on Linux
- Networking: a loopback-only MCP server behind a random secret path plus the tunnel you configure
| - Read/write access to the configured repo root (writable scope restrictable via --allow-path/--deny-path)
- Ability to invoke the chosen agent CLI (Claude/Codex/Gemini, etc.) and its underlying model account
- Local disk writes for run records, receipts, and share outputs (share/run-receipt., etc.)
- Execution of user-configured verifier commands (e.g. npm test)
|
|---|
| Risks and side effects | - Command execution runs with normal user privileges and is application-level containment, not an OS sandbox; same-user filesystem races remain possible
- With desktop control enabled, actions affect the entire machine, beyond your project folders
- The extension observes ChatGPT's web UI and multi-agent mode opens and types into extra ChatGPT tabs; this is not a documented public automation API and your account's terms, rate limits and permitted-use policies may apply
- Builds are not publisher-signed and macOS builds are not notarized, so SmartScreen, Gatekeeper or the browser will warn; verify SHA256SUMS.txt first
- Any public tunnel URL should be treated as a password; leaking it exposes the local MCP server
- Multi-agent mode and automatic generation consume additional ChatGPT or API quota
| - Verifier commands are executed; poorly chosen verifier commands carry security risk despite preflight policy checks
- Cost and token figures may be estimates (labeled with provenance) and should not be treated as settled accounting
- Receipt integrity must be verified before use as trustworthy evidence; missing rollback evidence is flagged as EVIDENCE_BOUNDARY
- Policy defaults come from martin.config.yaml and can be overridden by CLI flags; misconfiguration can loosen safety boundaries
|
|---|
| Supported clients | ChatGPT | Claude Code, Codex, Gemini CLI, Cursor, VS Code, GitHub Copilot, Continue |
|---|
| Tools | 12 | 0 |
|---|