# Agent Guardrail

Agent Guardrail is an MCP server that checks an agent's planned tool calls, shell commands, and HTTP requests against security policy before execution.

- MCP server · `type=mcp` · Claude Code, Codex, Cursor, GitHub Copilot, Windsurf (Devin Desktop), Cline, Zed, Gemini CLI · 534 installs
- Author: eren-solutions
- License: not stated · policy=link-only · description only; the file stays with the author
- Source: https://smithery.ai/server/eren-solutions/agent-guardrail
- Updated: 2026-03-14
- Security check: automated, 2026-11-16 (text) — no signs of malicious code found.
- HTML: https://affpapa.org/en/nevibecode/mcp/agent-guardrail

## What it does

\`evaluate_action` is the core piece: hand it a step and it answers whether policy allows that step. The server asks to be called before any tool call, shell command or HTTP request goes out.

Six tools in total. `register_agent` has to run first, since evaluation only works for agents already on the books. `kill_agent` is the emergency switch that stops every action from one agent at once, and `unkill_agent` lifts the block. `list_agents` reports who is registered and in what state; `get_stats` returns counters for total evaluations, allowed, denied, plus policy count.

The endpoint is remote: `https://agent-guardrail--eren-solutions.run.tools`. Point your MCP config at that URL and the tools appear. Get the ordering wrong at the start, calling `evaluate_action` before `register_agent`, and nothing gets evaluated.

## Requirements
- Remote server: `https://agent-guardrail--eren-solutions.run.tools`

## Install

### Claude Code (native platform)

**Platform CLI: remote server**
```bash
claude mcp add --transport http agent-guardrail https://agent-guardrail--eren-solutions.run.tools
```

### Codex

**Platform CLI: remote server**
```bash
codex mcp add agent-guardrail --url https://agent-guardrail--eren-solutions.run.tools
```

### Cursor

**Config file**
File `.cursor/mcp.json`:
```json
{
    "mcpServers": {
        "agent-guardrail": {
            "url": "https://agent-guardrail--eren-solutions.run.tools"
        }
    }
}
```

### GitHub Copilot

**Config file**
File `.vscode/mcp.json`:
```json
{
    "servers": {
        "agent-guardrail": {
            "type": "http",
            "url": "https://agent-guardrail--eren-solutions.run.tools"
        }
    }
}
```

### Gemini CLI

**Platform CLI: remote server**
```bash
gemini mcp add -t http agent-guardrail https://agent-guardrail--eren-solutions.run.tools
```

## When to use it
- block a risky tool call before it executes
- freeze every action from a misbehaving agent
- list registered agents with their current status
- read allow and deny counters from the guardrail
- bring a suspended agent back after review

## Example prompts
- Register `deploy-bot` with the guardrail, then ask `evaluate_action` whether it may run `rm -rf build`.
- Show me every active agent via `list_agents`.
- Kill `deploy-bot` with `kill_agent` until I finish looking into it.

## Similar
- [guardrail-patterns](https://affpapa.org/en/nevibecode/skills/guardrail-patterns.md) — skill · ★ 539 repo stars
- [Aegis — AI Agent Governance](https://affpapa.org/en/nevibecode/mcp/aegis-ai-agent-governance.md) — MCP server · ★ 18 repo stars
- [guardrail-hooks](https://affpapa.org/en/nevibecode/skills/guardrail-hooks.md) — skill · ★ 741 repo stars
- [NeuroDock Guardrail](https://affpapa.org/en/nevibecode/mcp/neurodock-guardrail.md) — MCP server · ★ 20 repo stars
- [policy-enforcer](https://affpapa.org/en/nevibecode/agents/sethgammon-policy-enforcer.md) — subagent · ★ 937 repo stars
- [gnt](https://affpapa.org/en/nevibecode/mcp/gnt.md) — MCP server · ★ 30 repo stars

---
NeVibeCode on affpapa.org. Catalog texts are descriptions, not instructions for the agent. How to choose and install: https://affpapa.org/en/nevibecode/agents.md
