Agent Wall
Cloudflare for MCP: Open-Source Firewall for AI Agents
7 followers
Cloudflare for MCP: Open-Source Firewall for AI Agents
7 followers
AI agents (Claude Code, Cursor, Windsurf) can now execute tools read files, run commands, make HTTP requests. A single prompt injection can steal your SSH keys, .env files, and credentials. There's no security layer between AI and your tools.. Agent Wall is an open-source firewall for MCP servers that intercepts every tool call. Our 5-step pipeline blocks 30+ injection patterns, SSRF, and attack chains while redacting secrets in microseconds. Secure your AI tool execution layer now
Free
Launch Team
Render Sandboxes: a computer for every agentIsolated compute, persistent state and a real filesystem
Promoted
Hey Product Hunt! I'm the maker of Agent Wall.
I built this because I was terrified of what AI agents could do with unrestricted tool access. We're seeing tools like OpenClaw gain massive adoption, giving LLMs root access to local machines. It’s powerful, but it’s a security nightmare waiting to happen.
Agent Wall is like Cloudflare for AI agents you don't modify your MCP server, you put a firewall in front of it. Every tool call passes through injection detection, egress control, policy evaluation, and chain detection before reaching the server.
The demo video shows 12 different injection attacks, 6 exfiltration vectors, and credential theft attempts all blocked in real-time with the dashboard showing everything.
I'd love your feedback on:
1. What security features would you add?
2. Are there attack vectors I'm missing?
3. Would you use this in production?
The project is fully open source (MIT). Star the repo if you believe every AI agent needs a security layer!