Discussion on Agent Authentication and Permission Models
jmppmj · reddit · 2026-07-14
The author explores how authentication and authorization should work when an agent needs to log into accounts, read email OTPs, use passkeys, or complete checkouts.
They argue that common solutions—putting credentials directly into the agent environment, integrating a secrets manager, or granting broad access—often give the agent far more permissions than the task requires. They are currently testing a more "user-controllable" model:
- The agent requests a specific account or piece of information via MCP
- The user approves the scope and duration of access on their phone
- Passwords are encrypted directly to the agent's public key
- Passkey assertions are signed on the phone
- Every access is logged in an audit trail
- Users can revoke permissions at any time
The author is implementing this mechanism into Decoy and wants to know if this account-scoped, time-limited permission model is actually valuable for consumer agents handling real user accounts, or if simply integrating an off-the-shelf secrets manager is the better approach.
More from coding & agent
- Codex helps build Valdiluce, an open-world game with climbing, gliding and gondolas — Dimillian · 2026-07-22
- HeyGen adds a media-sourcing skill for coding agents with 75k images and 10k tracks — HeyGen · 2026-07-22
- Agent search bottlenecks are now about variance, not raw latency — rohanpaul_ai · 2026-07-22
- LangSmith adds tracing for Pipecat, LiveKit, OpenAI Realtime, and Gemini Live — LangChain · 2026-07-22
- An MCP server signs every AI agent tool call into a verifiable Merkle chain — Funky_Chicken_22 · 2026-07-22
- Annotated transcript of a Claude Code team interview is now available — trq212 · 2026-07-22