MCP Guard
Real-time security firewall for AI agents
One line install, instant protection
MCP Guard is an inline security proxy between MCP clients and servers. Inspects all messages in real-time and blocks malicious tool calls before they reach the server.
Install globally:
npm install -g @aiclude/mcp-guardOr add to your MCP client config:
{
"mcpServers": {
"my-server-guarded": {
"command": "npx",
"args": ["-y", "@aiclude/mcp-guard", "--", "npx", "-y", "@your/mcp-server"]
}
}
}The Reality of MCP Threats
MCP has become the TCP/IP of AI agents. But existing security tools (WAF, SAST, DAST) don't understand MCP protocol.
92% Exploit Probability
With just 10 MCP plugins installed (VentureBeat)
30+ CVEs
MCP-related CVEs reported in Jan-Feb 2026 alone
OWASP MCP Top 10
Official threat classification published
3 Security Rule Engines
Full-spectrum inspection from tool definitions to tool calls to tool responses — no bypass path for attackers.
Tool Poisoning Detection
Detects hidden malicious instructions in tool descriptions: zero-width characters, homoglyphs, prompt injection, name spoofing
Argument Injection Detection
Blocks injection attacks in tool call arguments: SQL, Command, XSS, Path Traversal, Template — 21+ patterns
Data Exfiltration Detection
Detects sensitive data leaks in tool responses: credentials, system paths, stack traces
Advanced Detection Technology
Multi-layer detection engine that catches even invisible attacks.
Zero-Width Steganography
Detects 30+ invisible Unicode categories (U+200B, U+FEFF, etc.)
Homoglyph Normalization
47+ Cyrillic/Greek character mappings + NFKD normalization
Multilingual Prompt Injection
33+ injection phrases in English, Korean, Chinese, Japanese
Normalize-then-Match
Homoglyph → ASCII conversion → pattern matching to block character disguise bypass
Dual Protocol — Protection Everywhere
Local or remote, stdio or HTTP — seamless protection.
| Streamable HTTP | Legacy SSE | stdio | |
|---|---|---|---|
| Supported | ✅ POST/GET/DELETE | ✅ GET /sse, POST /messages | ✅ stdin/stdout |
Why MCP Guard?
The only solution that blocks in real-time at the MCP protocol level.
| Capability | MCP Guard | WAF | SAST | Manual Review |
|---|---|---|---|---|
| MCP Protocol Awareness | ✅ | ❌ | ❌ | △ |
| Real-time Blocking | ✅ | ✅ | ❌ | ❌ |
| Tool Poisoning Detection | ✅ | ❌ | ❌ | △ |
| Argument Injection Blocking | ✅ | △ | ✅ | △ |
| Data Exfiltration Detection | ✅ | △ | ❌ | △ |
| Multilingual Injection | ✅ | ❌ | ❌ | ❌ |
| Zero Config Deployment | ✅ | ❌ | ❌ | ❌ |
| Performance Impact | μs | ms | Offline | Offline |
Install today. Tomorrow may be too late.
Pre-block, not post-report. Zero performance impact. Zero configuration.
npm install -g @aiclude/mcp-guard