What is sidclaw-mcp-guard and how does it work?

It's a CLI proxy for MCP servers that applies YAML policies to tool calls: allow reads, hold writes for approval, block destructive ops. Runs locally with a dashboard.

How do I install MCP tool call guardrails?

npx sidclaw-mcp-guard@latest quickstart — generates policy.yaml, configures your client, starts the UI. Apache 2.0, no signup.

Does mcp-guard work with postgres and shell servers?

Yes, semantic patterns for SQL (read/write/destructive) and shell (safe/risky/destructive). Mix in one policy file.

🤖 AI & Machine Learning

Sidclaw's MCP-Guard: Finally, a Sanity Check for AI Agents Gone Wild

Picture this: your AI agent casually nukes a production database because 'hey, it seemed like a good idea.' Sidclaw's new MCP-guard slams on the brakes with human oversight.

theAIcatchup Apr 10, 2026 4 min read

Terminal demo of sidclaw-mcp-guard allowing SELECT, holding DELETE, blocking DROP TABLE on MCP server

⚡ Key Takeaways

Sidclaw-mcp-guard proxies MCP servers with YAML policies, human approvals, and audits—no code changes. 𝕏
Blocks destructive calls like DROP TABLE, holds writes, allows safe reads; works on SQL, shell, filesystems. 𝕏
Open source fix for agent risks, but layer it with proper perms; the firewall for AI tools. 𝕏

Published by

theAIcatchup

Community-driven. Code-first.

#AI agent guardrails #AI agent security #MCP servers #mcp-guardrails #sidclaw-mcp-guard #tool call approval #tool call security

Worth sharing?

Get the best Open Source stories of the week in your inbox — no noise, no spam.

Originally reported by Dev.to

⚡ Key Takeaways

The 60-Second TL;DR

theAIcatchup

Share this article

Worth sharing?

Related Stories

Karon's Sub-50ms Fetches Unlock the Real Web for AI Agents

Raspberry Pi 5 Unlocks Docker MCP Gateway: Linux Setup Sans Desktop

Slapped Governance on My Wild LangChain Agent in 30 Minutes—Microsoft's Toolkit Actually Works

Aerie's Agentic Workflows: Tools That Actually Think Before They Act

Stay in the loop