Your Agent sees a Handle. Never the Secret.
ClaudePass is a local secret manager built for AI coding agents. A command gets the real value at the moment it runs, injected by a local Broker into that one child process — the value never enters your Agent's Context: not the prompt, not a tool result, not a transcript.
How it works
Claude Code
cpass integrate claude installs an Intercept hook
(catches a pasted Secret before the model ever sees it), a
Command Policy hook (blocks commands that only exist to reveal
a Secret), and a skill that teaches Claude Code to reach for a
Handle instead of asking you for a value.
Codex
cpass integrate codex writes a delimited section
into AGENTS.md, so Codex learns the same Handles
and the same cpass run -- <command> pattern
— no separate integration to maintain.
MCP
cpass mcp speaks the Model Context Protocol over
stdio, exposing run_with_secrets and
capture as tools any MCP client can call — the
same Broker, Redaction, and Command Policy underneath.
Closed source. Documented like it isn't.
ClaudePass is closed-source software with a paid plan from day one — the same model 1Password and LastPass use, not open-core. Since you can't read the code yourself, trust has to come from documentation instead: exactly what the Vault, the Broker, both hooks, and the license check do and touch, every environment variable and on-disk path, and a plain no-telemetry statement beyond the license check.
Read the full security breakdown →
Free to try, simple to grow into
Free holds 3 Secrets — enough to wire up a real project end to end. Pro is $9.99/month for unlimited Secrets, verified entirely offline so a network blip never locks you out of a Secret you already have.