<div align="center">

🛡️ AIShield

Agent-Native AI Tool Security Scanner

Scan MCP servers, AI skills, GPTs, and prompts for security risks. 4-dimensional scoring. Certified badges. Guardrail MCP for auto-protection.

</div>

---

🚀 Quick Start

1. MCP Server (Claude Desktop / Cursor)

{
  "mcpServers": {
    "aishield": {
      "command": "npx",
      "args": ["aishield-mcp"]
    }
  }
}

2. Python SDK

pip install aishield

from aishield import AIShield

shield = AIShield()
result = shield.scan("https://github.com/modelcontextprotocol/servers")
print(result.overall_score)  # 85
print(result.badge_level)    # "gold"

3. Guardrail MCP (Auto-protection)

{
  "mcpServers": {
    "aishield-guardrail": {
      "command": "npx",
      "args": ["aishield-guardrail"],
      "env": {
        "AISHIELD_API_KEY": "your-key"
      }
    }
  }
}

Put aishield-guardrail FIRST in your config. It intercepts all tool installs and blocks unsafe ones.

4. GitHub Action

- uses: aishield/audit@v1
  with:
    api_key: ${{ secrets.AISHIELD_KEY }}
    fail_on_risk: true

---

🛡️ What AIShield Scans

Category	Detection
Tool Poisoning	Hidden adversarial instructions in tool descriptions
Prompt Injection	Malicious prompts that hijack agent behavior
Command Execution	child_process, subprocess, os.system
Data Exfiltration	Unauthorized network calls, telemetry
Credential Leaks	Hardcoded API keys, tokens, passwords
Dangerous APIs	File system, network, shell, database access
Supply Chain	Malicious dependencies, typosquatting
Code Quality	Error handling, input validation, docs

📊 Scoring

4-dimensional scoring (0-100):

Dimension	Weight	What it measures
Security	40%	Code vulnerabilities, dangerous APIs
Privacy	25%	Data exfiltration, telemetry
Quality	20%	Code quality, documentation
Performance	15%	Resource usage, efficiency

Badges: 🥇 Gold (≥85) | 🥈 Silver (≥70) | 🥉 Bronze (≥55)

---

📦 Packages

Package	Install	Description
aishield-mcp	npx aishield-mcp	MCP Server for Claude/Cursor
aishield-guardrail	npx aishield-guardrail	Guardrail MCP (auto-block unsafe tools)
aishield	pip install aishield	Python SDK
GitHub Action	uses: aishield/audit@v1	CI/CD integration
Claude Skill	Plugin install	Claude Code skill

---

🔌 API

Submit Audit

curl -X POST https://aishield.ai/api/v1/audit \
  -H "Content-Type: application/json" \
  -H "X-API-Key: your-key" \
  -d '{
    "tool_type": "mcp",
    "source_url": "https://github.com/user/repo",
    "name": "my-mcp-server"
  }'

Get Result

curl https://aishield.ai/api/v1/audit/{audit_id}

Get Badge

![AIShield](https://aishield.ai/api/v1/badge-name/your-tool-name)

Endpoints

Method	Endpoint	Description
POST	/api/v1/audit	Submit audit
GET	/api/v1/audit/{id}	Get audit result
GET	/api/v1/tools	List scanned tools
GET	/api/v1/stats	Global statistics
GET	/api/v1/badge-name/{name}	SVG badge
GET	/api/v1/pricing	Pricing info
POST	/api/v1/keys	Create API key
GET	/api/v1/health	Health check

---

💰 Pricing

Tier	Price	Scans/Day	Features
Free	¥0	5	Basic scanning, badge
Pro	¥29/month	200	Priority queue, batch scan
Enterprise	¥199/month	Unlimited	Custom rules, SSO, SLA
Pay-per-scan	¥1/scan	-	One-time

Get API key: https://aishield.ai/pricing

---

🏗️ Architecture

Agent (Claude/Cursor/Cline)
    ↓ installs MCP tool
Guardrail MCP intercepts
    ↓ calls AIShield API
AIShield Scanner
    ├── Static Analysis (30+ regex rules)
    ├── Dependency Analysis (npm/PyPI)
    ├── Secrets Detection
    └── Semantic Analysis (AI-powered)
    ↓ returns 4D score + badge
Agent shows result to user
    ✅ Approved → install
    🚫 Blocked → warn user

---

📈 Roadmap

• [x] MCP Server (stdio)
• [x] Python SDK
• [x] GitHub Action
• [x] Guardrail MCP
• [x] 4-dimensional scoring
• [x] Certified badges
• [ ] OWASP MCP Top 10 alignment
• [ ] Tool Poisoning deep detection
• [ ] Rug Pull detection (git diff monitoring)
• [ ] MCP Trust Framework (MTF) scoring
• [ ] Real-time handshake verification
• [ ] Batch scan 1000+ tools
• [ ] Industry security report

---

📄 License

MIT © AIShield