Five refusals our phone-calling MCP server enforces in code, not in the prompt
FunBus8049 · reddit · 2026-11-22
The author of UIRIX (a paid remote MCP server for Claude and ChatGPT, $9.99/mo) explains why instructions alone can't stop a phone-capable model from harm. Five rules are enforced as server-side refusals: no dialing without a recorded confirmation; tasks missing essentials return "needs input" before the call; the agent identifies itself as the user's AI assistant and never impersonates them; phone menus are listened to fully with a hold budget (3 min default, 10 max); and placing calls is a separately enabled permission. It never pays, finds businesses on its own, or calls for sales, collections or emergencies.
More from coding & agent
- Open-source MCP server brings Google Calendar access into Cursor via natural language — modelcontextprotocol · 2026-11-22
- Don't wire a chatbot to your personal WhatsApp: prompts make terrible firewalls — uriwa · 2026-11-22
- Sharing the Composer cache in GitHub Actions made PHPUnit integration tests 31% faster — DanielLockyer · 2026-11-22
- Google's Mirror Dimension: synthetic web for safe browser-agent testing — shashib · 2026-11-22
- Exposing Google/YouTube trending searches to agents via an MCP tool — Substantial-Tour-581 · 2026-11-22
- What hard stops do you put on your AI agents? A practical red-line list — Virtual_Arachnid7916 · 2026-11-22