A successful tool call isn't proof it happened: verifying agent writes with read-back checks
Prestigious-Web-2968 · reddit · 2026-10-15
The author kept hitting the same failure mode with MCP servers and agents: a write tool returns success, the host moves on — but the calendar event never landed, the draft isn't there, the booking has no record, and nothing errored.
Fix: verify writes with reads
- Map each write tool to its read counterpart (e.g. createevent → getevent); after the write, pass the new id to the read in the same request
- Read returns the object → hit; write "succeeded" but read is empty → miss. No read mapped? Just log the job ran and how long — don't invent a check
Two different hooks:
- MCP servers: wrap tool calls in-process with observe() recording tool name, latency and errors; hosts keep calling your URL directly, no proxying
- Personal agents (calendar/mail/booking via text): no public URL to watch, so Agent() wraps the dispatcher and looks for the matching list/get call after each job
The read check runs locally, never signs into anyone's Google, and only keeps job name, latency and found/miss. Already shipped as "Live Traffic" in the author's AgentStatus tool.
More from coding & agent
- Claude cut a 2.6-hour video export to 20 minutes — but only after being told to rebuild the server — byhewar · 2026-10-23
- 3 flight-booking rules my AI agent checks that I'd never do by hand — Efistoffeles · 2026-10-23
- OpenJet claims Qwen3.8-27B coding runs 1.9x faster than llama.cpp on one RTX 3090 — JeremyCMorgan · 2026-10-23
- Building an autonomous SEO agent in Hermes with three skills: programmatic pages, Keywords Everywhere, DataForSEO — EXM7777 · 2026-10-23
- Building a sports betting AI agent with onchain execution and human approval by default: architecture and lessons — Dry-Comfortable-2678 · 2026-10-23
- How do you audit AI agent decisions in regulated workflows? A field survey — Ok_Way_1993 · 2026-10-23