Plow Latch
Safely handle logins and credential entry for AI agents without putting passwords in model context
About Plow Latch
Plow Latch helps MCP-compatible AI agents complete browser actions that require login. Users can review the accounts, sites, files, and commands involved, then choose one-time or ongoing permission; Latch fills approved credentials directly in the browser, keeping them out of the agent's context and logs. The vault is encrypted on Mac, its key is protected by macOS Keychain, and no cloud copy is retained.
Best for
Best use cases
Have an agent complete actions on sites that require login
Fill approved credentials in the browser
Review accounts, sites, files, and commands before granting permission
Set ongoing permission for specific agents and sites
Features
Local encrypted vault
The vault is encrypted on Mac, with its key protected by macOS Keychain.
In-browser credential filling
Latch fills approved credentials directly in the browser; passwords never enter the agent's context or logs.
Request review and permission
Review accounts, sites, files, and commands, then choose one-time or ongoing permission.
Multi-agent MCP support
Supports Claude, Codex, OpenClaw, Hermes, and other MCP-compatible agents.
Pros
- Credentials are filled in the browser without entering the agent's context
- Passwords do not appear in logs
- Encrypted locally and protected by macOS Keychain
- Every request and decision is recorded