jcodemunch-mcp

Least-privilege permissions

• Create a dedicated GitHub token just for this tool instead of using your main account token. This limits damage if the token leaks—only code accessible to that restricted account can be viewed.
• Set token to "read-only" (no write, delete, or admin permissions). The tool only needs to look at code, never modify it.
• Regularly rotate credentials (replace the token every 30-90 days). This reduces the window an old, compromised token can be misused.
• Limit repository access by using GitHub organization controls to restrict which repositories the token account can see—exclude sensitive or production repos if possible.

Network isolation

• Run the connector on a dedicated machine or container separate from your main development environment. If compromised, the blast radius is smaller.
• Use a VPN or private network when connecting to the MCP server, especially if it communicates over the internet, to prevent token interception in transit.
• Disable unnecessary network access—if the tool only needs GitHub API access, firewall block all other outbound connections.

Monitoring

• Log all code retrieval requests (what files, when, by whom) and review them weekly for unexpected access patterns or unusual repository queries.
• Set alerts on your GitHub account for any new token creation or authentication failures, which may signal unauthorized access attempts.
• Track token usage via GitHub's audit logs to spot if your token is being called from unexpected locations or at unusual times.

Red flags to watch for

• Unexpected costs or API rate-limit spikes suggest the token may be used by unauthorized code or a compromised prompt.
• Requests for sensitive repositories you don't use—especially security, keys, or production systems—indicate possible manipulation or token leakage.
• Sudden changes in tool behavior, new error messages, or requests to upgrade permissions without explanation may signal a compromise or supply-chain attack.