We take security seriously. If you discover a security vulnerability in GraphRAG Codebase, please report it responsibly.

1. Do NOT open a public issue for security vulnerabilities.
2. Email the maintainers directly at the email address listed in the repository.
3. Include the following information:
   • Description of the vulnerability
   • Steps to reproduce
   • Potential impact
   • Any suggested fixes (optional)

• Acknowledgment: We will acknowledge receipt of your report within 48 hours.
• Assessment: We will assess the vulnerability and determine its severity.
• Fix Timeline: Critical vulnerabilities will be addressed as quickly as possible. We aim to release a fix within 7-14 days for critical issues.
• Disclosure: We will coordinate with you on public disclosure timing.

1. Use Strong Passwords: Never use default Neo4j passwords in production.

   # Change default password
   NEO4J_PASSWORD="your-secure-password"

2. Network Isolation: Run Neo4j on a private network, not exposed to the internet.

3. Enable Authentication: Always run Neo4j with authentication enabled.

1. Use Environment Variables: Never hardcode API keys in configuration files.

   export LLM_API_KEY="your-api-key"

2. Protect Configuration Files: Ensure .env has restrictive permissions.

   chmod 600 .env

3. Use Local Models When Possible: vLLM and Ollama provide local inference without sending data to external services.

The GraphRAG pipeline includes built-in protections against Cypher injection:

1. Schema Validation: Queries are validated against the graph schema.
2. Query Sanitization: User inputs are sanitized before query construction.
3. Parameterized Queries: All queries use Neo4j parameterized queries.

When indexing codebases, be aware that:

• Source code content may be stored in Neo4j
• Graph traversals can reveal code structure
• MCP tools expose query capabilities to connected agents

Ensure appropriate access controls are in place for sensitive codebases.

The MCP server exposes query tools to connected LLM agents:

• Run the MCP server on localhost only by default
• Enable authentication (MCP_REQUIRE_AUTH=true) for network deployments
• Use rate limiting (MCP_RATE_LIMIT_PER_MINUTE) to prevent abuse

1. Log Level: Use WARNING or higher in production to minimize data exposure.
2. Log Access: Restrict access to log files containing query patterns.

Security updates will be released as patch versions. Watch the repository releases to receive notifications of security updates.