Getting Started

Everything you need to install GuardClaw and start protecting your AI agents. No technical expertise required.

New to the terminal?

No worries. A terminal is just a text window where you type commands instead of clicking buttons. You'll copy a few short lines from this page, paste them in, and press Enter. We walk through every step below, starting with how to open a terminal on your machine.

Already comfortable with the command line? Skip to installation.

Prerequisites

You'll need a Terminal to run installation commands. Here's how to open one:

macOS

Press Cmd + Space, type "Terminal", and press Enter.

Linux

Press Ctrl + Alt + T to open Terminal.

Windows

Press Win + X, then select "Terminal" or "PowerShell".

On macOS or Linux, we recommend installing via Homebrew. If you don't have it, install it first:

/bin/bash -c "$(curl -fsSL https://raw.githubusercontent.com/Homebrew/install/HEAD/install.sh)"

Already have Homebrew? Check with brew --version. If it prints a version number, you're good.

On Windows, no package manager is required — the PowerShell installer handles everything. See installation options below.

Installation

Install GuardClaw

Paste this into your Terminal and press Enter:

brew install TakeInterestInc/tap/guardclaw

That's it. Global protection activates automatically. Every Claude Code session on your machine is now protected with no additional setup.

Verify it's working

Run these commands to confirm GuardClaw is installed and healthy:

guardclaw version    # shows the installed version
guardclaw doctor     # checks hooks, binary, and all defense layers

Upgrade to the latest version

brew update && brew upgrade guardclaw

Upgrades automatically update global hooks. New security patterns and detection rules propagate immediately.

Uninstall

To completely remove GuardClaw and all its configuration:

brew uninstall guardclaw                           # remove the binary
guardclaw init claude-code --global --uninstall    # remove global hooks (run before uninstall)

To start fresh (e.g. after a failed install or checksum error):

brew uninstall guardclaw 2>/dev/null; brew untap TakeInterestInc/tap 2>/dev/null
brew tap TakeInterestInc/tap && brew install TakeInterestInc/tap/guardclaw

This removes the cached tap, re-fetches it, and reinstalls cleanly.

Windows (PowerShell)

Open PowerShell and run:

irm https://get.guardclaw.dev/install.ps1 | iex

Downloads and installs GuardClaw for Windows. Requires PowerShell 5.1 or later.

Shell installer (all platforms)

Works on macOS, Linux, and WSL:

curl -fsSL https://get.guardclaw.dev/install.sh | sh

An alternative to Homebrew if you prefer a standalone installer. Downloads the correct binary for your platform automatically.

Platform Setup

GuardClaw protects all sessions globally after install. For per-project configuration or to set up the MCP server, use the commands below.

Claude Code

Global protection is automatic after brew install. To add per-project MCP and hooks:

cd your-project-folder
guardclaw init claude-code

What this creates:

.mcp.json — Registers the GuardClaw MCP server so Claude Code routes tool calls through policy enforcement
.claude/settings.local.json — Hook configuration that intercepts Read, Write, Bash, and other tools before they execute

Per-project settings override global defaults. Restart Claude Code after running this command.

Start a session: Type claude in your terminal to launch Claude Code with GuardClaw protection active.

Claude Desktop

Set up MCP server integration for Claude Desktop:

guardclaw init claude-desktop

Adds GuardClaw as an MCP server to your Claude Desktop configuration at ~/Library/Application Support/Claude/claude_desktop_config.json.

Start a session: Open Claude Desktop from your Applications folder. GuardClaw protection is active automatically.

Cursor

Set up GuardClaw for Cursor:

guardclaw init cursor

Configures MCP server integration and policy enforcement for Cursor's AI features.

Start a session: Open Cursor normally. GuardClaw protection is active automatically for AI-powered actions.

OpenClaw

Integrate GuardClaw with OpenClaw:

guardclaw init openclaw

Auto-detects your OpenClaw installation and configures policy enforcement.

Start a session: Type openclaw in your terminal to launch OpenClaw with GuardClaw protection active.

Run guardclaw init with no arguments to see the full list of supported platforms and options.

How It Works

GuardClaw protects your AI agents at two levels:

Hooks (automatic)

Every tool call (Read, Write, Bash, etc.) is intercepted before it executes. GuardClaw checks the action against 1,743 detection patterns and blocks anything suspicious.

MCP server (per-project)

When configured, GuardClaw runs as an MCP server that your AI tool connects to. All tool calls are routed through policy enforcement before reaching your system.

No LLM in the security path. Decisions are made by rules, not predictions. Your data stays in your infrastructure.

Usage

Once installed, GuardClaw works automatically in the background. You can also use it directly:

guardclaw status               # check active protection
guardclaw doctor               # run health checks
guardclaw run -- <command>     # wrap a command with defense layers

Most users never need to run these. GuardClaw works silently, only surfacing when it blocks a suspicious action.

Cloud Dashboard (Optional)

The optional cloud dashboard provides audit history, agent monitoring, and team management. Connect from the CLI:

1. Connect your workspace

guardclaw connect --api-key <your-key> --workspace <your-workspace>

2. Sync receipts

Every policy decision GuardClaw makes is recorded locally as a receipt, an append-only, SHA-256 hash-chained audit log. To upload your audit trail to the cloud dashboard:

guardclaw sync

Receipts include what was checked, the decision (allow, deny, or escalate), and a tamper-evident hash chain you can verify at any time:

guardclaw status               # view current config and active agents

3. Check status

After connecting, verify everything is linked:

guardclaw status

The cloud dashboard is optional. GuardClaw works fully offline with no external dependencies. Receipts are always stored locally regardless of whether you sync them.

Troubleshooting

"command not found: guardclaw"

Close and reopen your Terminal, then try again. If it still doesn't work, run brew install TakeInterestInc/tap/guardclaw to install.

"SHA256 mismatch" or checksum error

Clear the cached tap and reinstall:

brew untap TakeInterestInc/tap && brew tap TakeInterestInc/tap && brew install TakeInterestInc/tap/guardclaw

"command not found: brew"

You need to install Homebrew first. See the Prerequisites section above.

GuardClaw not appearing in your AI tool

Run brew update && brew upgrade guardclaw to ensure you have the latest version, then completely quit and relaunch your AI tool.

Hooks pointing to dead path

Run guardclaw doctor to diagnose. The gate auto-repairs stale hooks on the next tool call, or run guardclaw init claude-code --global to reinstall global hooks manually.

"low max file descriptors" or unexpected error on launch

Increase the file descriptor limit before starting Claude Code:

ulimit -n 10240 && claude

To make this permanent, add ulimit -n 10240 to your ~/.zshrc (macOS) or ~/.bashrc (Linux).

Safe operations being blocked

GuardClaw has configuration options to adjust enforcement for local development. Run guardclaw doctor for specific guidance.

Need help? Contact us at security@takeinterest.ai.

Frequently Asked Questions

Does GuardClaw send my data anywhere?

No. Everything runs locally on your machine by default. Your code, prompts, and tool calls never leave your infrastructure. The cloud dashboard is entirely optional. If you choose to sync receipts, only the audit log (what was checked and the decision) is uploaded, not the content of your files or conversations.

Does GuardClaw use AI or an LLM to make security decisions?

No. All decisions are deterministic and rule-based. Pattern matching, policy enforcement, and anomaly detection are all computed locally with no model in the security path. This means decisions are fast, predictable, and auditable.

What platforms does GuardClaw support?

Claude Code, Claude Desktop, Cursor, OpenClaw, and any MCP-compatible client. GuardClaw also supports HTTP mode for custom agent integrations. See Platform Setup above for details.

Will GuardClaw slow down my AI tool?

No noticeable difference. Policy checks run in microseconds against compiled patterns. GuardClaw intercepts tool calls before execution, so the overhead is negligible compared to the time the AI model spends generating a response.

What are receipts?

Every time GuardClaw evaluates a tool call, it creates a receipt: a tamper-evident record of what was checked and the decision. Receipts are stored locally in an append-only, SHA-256 hash chain. You can verify the integrity of the chain at any time. If you connect to the cloud dashboard, you can sync receipts for team-wide audit visibility.

How do I completely remove GuardClaw?

Remove the global hooks first, then uninstall the binary:

guardclaw init claude-code --global --uninstall    # remove global hooks
brew uninstall guardclaw                           # remove the binary

See the Uninstall section above for a full fresh-start procedure.

Can I customize which operations are blocked?

Yes. GuardClaw supports multiple strictness levels (paranoid, strict, balanced, permissive) and custom allow rules. Run guardclaw doctor for guidance on adjusting enforcement for your workflow.

Is GuardClaw free?

The public beta is free to use. See the GuardClaw page for current pricing details.

How do I report a security issue?

Email security@takeinterest.ai. Security reports are treated with priority. See our Security page for our disclosure policy.

Understanding GuardClaw's scope

GuardClaw is in public beta. It is actively maintained, regularly updated, and used in real workflows. That said, we want to be upfront about what it does and what it does not do.

What GuardClaw does

GuardClaw enforces deterministic, rule-based controls on what AI agents can execute on your machine. It intercepts tool calls before they run, checks them against 1,743 detection patterns across multiple attack categories, and blocks actions that violate policy. Everything runs locally. No data leaves your infrastructure unless you opt into the cloud dashboard.

What GuardClaw does not do

GuardClaw is not a replacement for network firewalls, endpoint protection, identity management, or access control systems. It does not scan, sandbox, or modify the AI model itself. It controls what the agent can do, not what the model thinks. Pattern detection covers known attack shapes. Novel techniques that do not match existing patterns may not be caught until the next update ships. No security tool catches everything, and we will not claim otherwise.

Public beta means:

The product is stable enough for real use and we stand behind it. It also means we are still shipping improvements, hardening edge cases, and expanding coverage. You may encounter rough edges. When you do, we want to hear about them.

GuardClaw gets better as the community uses it. Detection patterns grow, policy defaults sharpen, and edge cases get covered faster when more teams are running it in real workflows. Early adopters shape what this becomes.

Report issues or feedback to security@takeinterest.ai. Security-related reports are treated with priority.

GuardClaw Terms of Service Privacy Notice Security