What is Agent Wall?

Agent Wall is a two-way security firewall for AI agents. It intercepts MCP (Model Context Protocol) tool calls going in and scans server responses coming out, enforcing YAML-based policies and blocking dangerous operations before they happen.

The Problem

AI agents can now execute tools — read files, run shell commands, query databases, make HTTP requests. Without guardrails, a single prompt injection can:

1. Read your SSH keys, .env files, credentials
2. Exfiltrate data via curl, wget, or DNS tunneling
3. Execute arbitrary shell commands with your permissions
4. Leak secrets in tool responses — API keys, tokens, private keys flow right back to the LLM

The Solution

Agent Wall sits between the MCP client (Claude Code, Cursor, etc.) and the MCP server, enforcing a YAML policy on every tool call:

MCP Client  ←→  Agent Wall Proxy  ←→  MCP Server
                      ↕
               agent-wall.yaml
           (rules + responseScanning)

Think of it as Cloudflare for AI agents — transparent, zero-config, and powerful.

Key Features

• Two-way firewall — Inspects tool calls going in AND responses coming out
• Policy enforcement — YAML-based rules with glob matching and argument inspection
• Response scanning — Detects leaked API keys, tokens, private keys, PII in server responses
• Three request actions — allow, deny, prompt (ask human for approval)
• Three response actions — pass, redact, block
• Zero config — Sensible built-in defaults that protect immediately
• Transparent proxy — No SDK changes, no code modifications needed
• Audit logging — Structured JSON logs with sensitive value redaction
• Rate limiting — Global and per-rule limits
• Security scanning — Find unprotected MCP servers in your config

How is it different?

Unlike traditional API gateways, Agent Wall:

• Works at the MCP protocol level (JSON-RPC over stdio)
• Understands tool semantics — not just HTTP verbs
• Scans both directions — requests AND responses
• Provides human-in-the-loop approval for dangerous operations
• Redacts secrets from responses before the LLM sees them
• Requires zero changes to your MCP servers
• Is designed specifically for AI agent security

Works with any MCP server

Agent Wall is protocol-level, not application-level. It works with every MCP server that uses stdio transport — official servers, third-party servers, or your own custom ones. There are 1000+ MCP servers in the ecosystem and growing; Agent Wall protects them all without any code changes.

# Official servers
agent-wall wrap -- npx @modelcontextprotocol/server-filesystem /path
agent-wall wrap -- npx @modelcontextprotocol/server-github

# Third-party servers
agent-wall wrap -- npx @supabase/mcp-server-supabase
agent-wall wrap -- npx @stripe/mcp-server-stripe

# Your own servers
agent-wall wrap -- node my-custom-mcp-server.js
agent-wall wrap -- python my_mcp_server.py

Works with any MCP client

Any MCP client that spawns servers via stdio can use Agent Wall:

• Claude Code / Claude Desktop
• VS Code / GitHub Copilot (MCP support since v1.99)
• Cursor
• Windsurf (Codeium)
• Cline, Continue.dev
• Any custom MCP client