Agentguard

Overview

Agentguard is a Model Context Protocol server that acts as a network-egress firewall for agent tools. It evaluates URLs against a declarative policy, giving you control over where your AI agents can make outbound requests. Integrated directly into your AI assistant, Agentguard exposes its URL-checking capability as a tool your model can invoke.

The server runs locally on your machine via the npm package @mukundakatta/agentguard-mcp, communicating over stdio with no external dependencies or credentials required. Version v0.1.0 is available now under the MIT license, with full source code hosted at MukundaKatta/agentguard-mcp on GitHub.

Agentguard belongs in the AI & Agents category and works with Claude Desktop, Claude Code, Cursor, VS Code, Windsurf, Zed, and any MCP-compatible client. If you maintain this server, claim your listing to earn the Verified badge and keep your information accurate.

Installation

// claude_desktop_config.json
{
  "mcpServers": {
    "agentguard": {
      "command": "npx",
      "args": [
        "-y",
        "@mukundakatta/agentguard-mcp"
      ]
    }
  }
}