Skip to main content
Documentation▾

Introduction

Unclaw is an open-source security proxy for AI agents. It sits between your agent and the internet, intercepting all network traffic. Agents never touch your API keys directly — Unclaw injects them into outgoing requests on the fly.

The Problem

Your AI agent has every API key in plaintext. It talks to GitHub, Slack, Anthropic, and dozens of other services. You can't see what it's doing, what it costs, or where your credentials end up. One prompt injection and your secrets are exfiltrated.

What Unclaw Does

  • Secret injection. Agents use placeholders. Unclaw replaces them with real credentials before forwarding. Your agent never sees the actual key.
  • Full visibility. Every request is logged — method, host, headers, body, response, latency, LLM token usage and cost.
  • Works with any agent. Claude Code, Codex, OpenClaw, or your own scripts. No code changes needed.
  • Plugin system. Built-in support for Anthropic, OpenAI, GitHub, Slack, and more. Write your own plugins for custom services.

How It Works

Unclaw runs as a local proxy on your machine. On macOS, it uses a Network Extension to transparently intercept traffic from specific processes. On Linux, it uses a WireGuard tunnel inside a network namespace. Either way, the agent's traffic is routed through Unclaw without the agent knowing.

Agent  -->  Unclaw Proxy  -->  api.anthropic.com
              |
              +-- injects API key
              +-- logs request
              +-- tracks token usage

Open Source

Unclaw is fully open source (MIT). You can run it locally with no external dependencies — all data stays on your machine in a SQLite database.

Next Steps