40 MCP tools connected: which ones would you never let your agent call without approval?
JuicerSocial · reddit · 2026-10-22
A discussion prompt: an AI agent has 40 MCP tools configured — reading databases, searching files, creating tickets, updating records, sending messages, deleting data, triggering workflows. Now someone suggests removing the approval prompts. Where do you draw the line?
Deleting data and moving money are obvious, but the poster asks about subtler cases: updating a CRM field, sending a Slack message, modifying a calendar event, or triggering an automation that chains into five more — seemingly harmless tools with significant downstream effects.
The author asks how people handle this: tool-level permissions, approval gates, sandboxing, or keeping certain capabilities entirely unavailable — and what the most innocent-looking MCP tool is that you'd never let an agent call autonomously.
More from coding & agent
- Classic 'too many tabs' meme applied to coding agents: send help — mrdrozdov · 2026-10-22
- YC F26 startup Latent reads model internals to catch hallucinations at 530x lower cost — ycombinator · 2026-10-22
- We need an L0.5 for continual learning: agent state stacks up in 4 layers — willcb · 2026-10-22
- DuckDB + Hugging Face Amsterdam workshop: build an AI feedback pipeline over 2.1M Amazon reviews — lhoestq · 2026-10-22
- Veracode report: AI now writes half of committed code but only 56% pass security checks — WeldPond · 2026-10-22
- Will Brown: Continual Learning in the Harness Is Like a Computer Without RAM — willcb · 2026-10-22