Personal AI ops agent starter kit — Cursor edition. Provides Discord slash commands for agent control, scheduled cron job execution, and Claude model management.
This server exposes 7 tools via Discord slash commands, but lacks proper MCP tool registration. Tool definitions are inferred from Discord command handlers rather than explicitly registered with the MCP protocol. No input schemas are visible in the source code, parameters are extracted via Discord.js interaction.options methods, not declared as JSON Schema. Descriptions exist but are minimal (10-50 chars). No output schemas documented. Tools mix concerns (e.g., 'herc' sends prompts AND runs Claude). Error handling is basic (try/catch with generic messages). Security issues: Discord bot token exposed in env config; no permission gates; no audit logging.
List all enabled cron jobs
Tail the log for a cron job
Manually trigger a cron job
Send an arbitrary prompt to your agent
Show or set the active Claude model
Post a message to a Discord channel via bot token
Check agent health — uptime, memory, cron status
No JSON Schema input definitions visible. Tools are Discord command handlers, not MCP-registered tools with explicit schemas. Cannot verify parameter types, constraints, or required fields.
Tool descriptions are too short (10-50 chars). Descriptions like 'Check agent health' and 'Show or set the active Claude model' lack context on WHEN to use, WHAT happens, and any prerequisites. LLMs cannot reliably select tools with such minimal guidance.
No output schemas documented. Tools return free-text or unstructured Discord messages. LLMs cannot plan downstream calls or extract structured data reliably.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 31 | 2026-07-28+ | v2 |
Discord bot token exposed in .cursor/mcp.json via DISCORD_BOT_TOKEN env var. Credentials must never appear in config files or be passed as tool parameters. Use server-side secret injection.
No permission gates or audit logging. Tools like 'cron run' and 'model' modify state without verifying caller authority or logging who did what. No compliance trail for agent-initiated actions.