agentx-mcp Proxy Blocks Destructive AI Agent Calls and Guides Self-Recovery
A new open-source tool called agentx-mcp acts as a lightweight stdio proxy that intercepts every tool call made by AI agents running through MCP servers before execution. The proxy deterministically blocks dangerous operations such as DROP TABLE queries, unscoped deletes, shell teardown commands, and SSRF attempts without requiring an API key or any external service. When a harmful call is blocked, the proxy returns a structured coaching error that describes what was unsafe and suggests a safer alternative, allowing the agent to revise and retry on its own. This self-correction loop means an autonomous agent run can survive a blocked call and still complete its task, rather than failing on a hard error. The tool is installed via pip and configured with a single line change in the mcp.json file, making it compatible with any MCP-speaking client such as Claude Code or Cursor.
This is an AI-generated summary. ShortSingh links to the original source for the complete article.
Discussion (0)
Log in to join the discussion and vote.
Log in