AgentGuards plugin for Claude Code

LLM security guardrails for Claude Code in one install: jailbreak and prompt-injection detection, web-content scanning, data-exfiltration blocking, and destructive-command authorization.

Enforcement is configurable: fail-closed by default for strict security, or switch to fail-open (availability-first) with a single environment variable (AGENTGUARDS_FAIL_OPEN=true).

This plugin bundles:

• the AgentGuards MCP server (check_input, authorize_action, validate_output, evaluate_policy, health_check),
• enforcing hooks — UserPromptSubmit input scanning, PreToolUse Bash authorization, and PostToolUse web-content scanning/redaction,
• the AgentGuards security instructions (the guardrails skill).

Install

/plugin marketplace add alelaguard/agentguards-plugins
/plugin install agentguards-claude@agentguards

Then provide your API key (get one at https://agentguards.co/dashboard/keys) so both the MCP server and the hooks can authenticate:

export AGENTGUARDS_API_KEY=ag_your_token_here

Add that line to your shell profile (~/.bashrc, ~/.zshrc, …) and restart Claude Code. Or just run /agentguards:setup and it will walk you through it.

Alternative: npm install @agentguardsco/claude-plugin. Fetches these same files for programmatic use (pinned versions, CI, custom tooling) — it does not register with Claude Code on its own; use /plugin install above for that.

Commands

• /agentguards:setup — set your API key and verify everything is wired up.
• /agentguards:status — report whether the guardrails are active and healthy.

Configuration

| Variable | Required | Default | Purpose | |---|---|---|---| | AGENTGUARDS_API_KEY | yes | — | Your ag_ token. Drives both the MCP header and the hooks. | | AGENTGUARDS_URL | no | https://prod.agentguards.co | Override only for a self-hosted instance. | | AGENTGUARDS_FAIL_OPEN | no | false | Hooks fail closed by default (block when the service is unreachable). Set true to allow on error. |

How it works

The hooks call the AgentGuards REST API on every prompt, before every Bash command, and after every web fetch — blocking or redacting when AgentGuards flags a risk. The MCP tools let Claude cooperatively check inputs and authorize actions as described in the bundled guardrails skill.

Learn more at https://agentguards.co.