This is a hardened fork of CodeGraphContext with additional security measures.

The add_code_to_graph tool now validates paths to prevent:

• Indexing of sensitive files (.ssh, .env, credentials, etc.)
• Path traversal outside allowed directories
• Symlink escape attacks

Configuration:

# Allow specific directories (colon-separated)
export CGC_ALLOWED_ROOTS="/home/user/projects:/home/user/workspace"

# Default: current working directory only

Blocked patterns:

• .ssh, .gnupg, .aws, .kube (credentials)
• .env, .envrc, credentials, secrets
• *.pem, *.key, id_rsa, id_ed25519
• /etc/shadow, /etc/passwd

The execute_cypher_query tool now:

• Validates query prefix (MATCH, WITH, RETURN only)
• Blocks write operations (CREATE, MERGE, DELETE, etc.)
• Supports parameterized queries
• Normalizes Unicode to prevent bypass

Parameterized queries (recommended):

# Instead of string interpolation
query = "MATCH (n) WHERE n.name = $name RETURN n"
params = {"name": user_input}
result = execute_cypher_query(db, cypher_query=query, params=params)

For security issues in this hardened fork, open an issue with the security label.

For upstream CodeGraphContext security issues, see: https://github.com/CodeGraphContext/CodeGraphContext/security

Full security audit: memory/research/codegraphcontext-security-audit.md

Verdict: Safe for local use with trusted codebases.

Original upstream security policy follows below.

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

DO NOT create a public GitHub issue for security vulnerabilities.

Instead, please:

1. Email your findings to security@codegraphcontext.dev
2. Include a detailed description of the vulnerability
3. Provide steps to reproduce if possible
4. Allow us 90 days to respond before public disclosure

We provide security updates for the latest major version only.

• The MCP server listens on stdin/stdout only (no network exposure)
• No authentication is required (local process communication)
• Only run the MCP server from trusted parent processes

• Credentials are stored in environment variables or .env files
• Never commit credentials to version control
• Use environment-specific credential files

• The indexer can read any file the process has access to
• Only index trusted codebases
• Consider running in a container for untrusted code

• The execute_cypher_query tool is restricted to read-only operations
• Write operations (CREATE, MERGE, DELETE) are blocked
• Custom deployments may need additional restrictions