discord-agent-mcp

The project uses TypeScript and Zod for strong type and input validation, significantly reducing common injection vulnerabilities. Discord.js's permission handling is correctly leveraged for all operations, enforcing least privilege. Configuration (e.g., DISCORD_TOKEN) is handled via environment variables, with clear instructions against committing secrets. The `send_message_with_file` tool allows specifying an absolute file path; while validated for existence, a malicious AI could potentially exfiltrate arbitrary files if the bot's underlying OS permissions allow it. However, the Dockerfile includes `runAsNonRoot: true` and `allowPrivilegeEscalation: false`, which mitigates this risk by limiting file system access. No `eval` or similar dangerous functions were found.