Should your AI agent write the IBAN validator itself, or call a versioned MCP tool instead?
sandeepK1125 · reddit · 2026-11-17
The Apyhub team (disclosure: their MCP server is the product in question) frames a real engineering dilemma: ask an agent to "validate these IBANs" or "check this VAT number" and it confidently improvises a small function each time — different implementations, inconsistent edge-case handling, zero testing.
Their answer: for tasks with one right answer (IBAN/VAT checks, sanctions screening, PDF conversion, invoice extraction), expose each as a versioned service behind MCP so the agent calls the same tested code every run, across models. The tradeoffs are an extra network call and dependency — and the harder problem of giving the model a good reason to pick the tool over its own code, since models are overconfident they can write a checksum function.
They open two questions for practitioners: which tasks you let agents code vs. force through tools, and how to stop the "just quickly write it" reflex when a tool exists.
More from coding & agent
- Hermes App Screenshot Exposes Meta's Unreleased muse-wonder-1.0 Model with 1M Context — Teknium · 2026-11-17
- Leaderboard ranks agent harnesses for open models: Claude Code, Codex, Pi, OpenCode, Hermes and more — AxSaucedo · 2026-11-17
- Homoglyph attack bypasses email-agent keyword filters: one swapped letter defeats "ignore previous instructions" blocklist — Spacesh1psoda · 2026-11-17
- poteto on autonomous agents: 6 months of infra investment led to thousands of PRs while sleeping — 4310sy · 2026-11-17
- Tencent Hunyuan's Accelerate Bench tests 16 coding agents on 32 CPU optimization tasks — Tencent-Hunyuan · 2026-11-17
- SWE-CoWork: coding agents pass 63% of tests but complete fewer than 8% of tasks — Tokenwave-AI · 2026-11-17