MEDUSA integrates with popular AI-powered IDEs to provide intelligent security scanning with automatic false positive detection.

Run this command in your project root:

medusa init

This creates all necessary configuration files for detected IDEs.

project/
├── CLAUDE.md                     # Project context with FP handling
├── .claude/
│   ├── agents/medusa/
│   │   └── agent.json           # Agent configuration
│   └── commands/
│       ├── medusa-scan.md       # Scan command
│       └── medusa-install.md    # Install command

Run a security scan:

/medusa-scan

Quick scan (cached):

/medusa-scan --quick

Install missing tools:

/medusa-install

Claude Code reads CLAUDE.md and automatically:

1. Identifies likely false positives (B404, B603, B607, B101)
2. Highlights real security issues
3. Suggests creating .bandit config for project-wide FP handling
4. Provides fix suggestions for real vulnerabilities

You: /medusa-scan

Claude: Running MEDUSA security scan...

Found 45 findings. After intelligent triage:

**False Positives (43):** Subprocess usage in CLI tool - expected
**Real Issues (2):**
1. HIGH: api/auth.py:156 - SQL query built with string concatenation
   Fix: Use parameterized query with SQLAlchemy
2. MEDIUM: config.py:23 - Debug mode enabled
   Fix: Use environment variable for DEBUG setting

Shall I create a `.bandit` config to suppress the FPs?

project/
├── GEMINI.md                     # Project context
├── .gemini/
│   └── commands/
│       ├── medusa-scan.toml     # Scan command
│       └── medusa-install.toml  # Install command

# In Gemini CLI
/medusa-scan
/medusa-install

The /medusa-scan command includes a 4-step triage process:

1. Identify False Positives - Known FP patterns
2. Identify Real Issues - Actual vulnerabilities
3. Report Summary - Filtered results
4. Handle FPs - Suggest .bandit config

project/
├── .github/
│   └── copilot-instructions.md  # Copilot context

Copilot reads copilot-instructions.md and applies the rules when:

• Suggesting code completions
• Reviewing security scan results
• Answering questions about security

The file teaches Copilot:

• Security patterns to avoid (SQL injection, XSS, etc.)
• How to recognize false positives
• When to suggest running medusa scan .
• How to handle FPs with .bandit config

project/
├── AGENTS.md                     # Codex context

• Security requirements
• Common FP patterns with explanations
• Table of Bandit rules and when they're FPs
• How to create .bandit config

Cursor uses both Claude Code and MCP (Model Context Protocol):

project/
├── CLAUDE.md                     # Shared with Claude Code
├── .claude/                      # Shared commands
├── .cursor/
│   └── mcp.json                 # MCP server config

The mcp.json configures MEDUSA as an MCP server:

{
  "mcpServers": {
    "medusa-security": {
      "command": "medusa",
      "args": ["mcp-server"]
    }
  }
}

Edit the generated context file (e.g., CLAUDE.md) to add:

## Project-Specific Security Rules

- All API endpoints must validate JWT tokens
- Database queries must use the ORM, never raw SQL
- File uploads must be scanned for malware

MEDUSA's init command won't overwrite existing context files. Your customizations are safe.

To regenerate (and overwrite):

medusa init --force

Edit .claude/agents/medusa/agent.json:

{
  "triggers": {
    "file_save": {
      "enabled": true,
      "patterns": ["*.py", "*.js", "*.ts"]
    }
  },
  "settings": {
    "auto_scan": true,
    "inline_annotations": true
  }
}

Set in .medusa.yml:

ide:
  claude_code:
    enabled: true
    auto_scan: false

• Use /medusa-scan --quick for fast feedback during development
• Ask Claude to explain findings: "What does B602 mean?"
• Request fixes: "Fix the SQL injection in auth.py"

• The scan command includes detailed triage steps
• Gemini will suggest .bandit config when seeing many FPs

• Copilot applies security rules when suggesting code
• After completing code, it may suggest running a scan
• Ask: "Is this code secure?" for instant review

• Works like Claude Code with additional MCP capabilities
• Can run scans through MCP protocol
• Shares context files with Claude Code

# Regenerate IDE files
medusa init --force

# Check files were created
ls -la .claude/ .gemini/ .github/ .cursor/

Ensure the context file is in the project root:

• CLAUDE.md (not .claude/CLAUDE.md)
• GEMINI.md (not .gemini/GEMINI.md)

Create .bandit config as described in Handling False Positives.

# Verify MEDUSA is installed
medusa --version

# Verify it's in PATH
which medusa

• Handling False Positives
• MEDUSA Configuration
• Claude Code Best Practices
• Gemini CLI Custom Commands