Per-agent short-lived tokens for MCP servers: revoke one agent in ~1 minute
zach-benson · reddit · 2026-11-09
A developer is seeking 1-2 testers for a per-agent token scheme for MCP servers holding notes, memory, documents, or meeting data. The pain point: users connect Claude, ChatGPT, Cursor, and their own agents to the same account, and you can't cut off one agent without signing the user out everywhere.
The scheme: each agent gets its own short-lived token carrying the user, an agentid, and allowed tools (read-only to start). Revoking one agent is rejected within 1 minute while others keep working. Integration costs 30 minutes — a snippet plus two settings, observe-only (tokens are checked and logged, nothing blocked), existing auth untouched, fully reversible. Benefits: per-agent visibility in logs (e.g., "Cursor called searchnotes 40 times"), a one-agent kill switch on read tools, and a write-up including failures. The author is building an open-source library in this area.
Related event: Developer Seeks Pilots for Per-Agent Short-Lived Tokens on MCP Servers(2 posts)→
More from coding & agent
- Solo dev's infinitely evolving Claude-built Backrooms game hits 35K players in 36 hours — minchoi · 2026-11-09
- RentAHuman lets AI agents hire real people for physical-world tasks via MCP API — conitzer · 2026-11-09
- Jane Street open-sources /socratic-tutor: an agent skill that teaches via questioning, not lecturing — daniel_mac8 · 2026-11-09
- Fly-by-Code: coding agents pilot drones via active visual and physical feedback loops — jbhuang0604 · 2026-11-09
- Unreal Engine 5.9 preview is the first UE release built primarily with agentic workflows — nicolascraske · 2026-11-09
- YC-backed Floworks launches Eniac, an API for verified large-scale web data extraction — ycombinator · 2026-11-09