MCP Tool Gateways Should Fail Loudly: A Plausible Answer From a Missing Dependency Is the Real Bug
Emergency-Tutor-7 · reddit · 2026-11-23
A practitioner's write-up of fault-tolerance patterns for MCP tool gateways, backed by an MIT-licensed reference implementation (scientific-toolkit-mcp on PyPI). The core insight: when an agent calls a tool that isn't actually installed, the dangerous outcome isn't an error — it's a plausible fabricated answer the model can't distinguish from a real result, which surfaces as an untraceable wrong conclusion steps later.
Patterns: probe dependencies at call time, not startup; return machine-branchable statuses like NOTINSTALLED instead of vague errors; keep upstream packages as independent repos rather than vendoring (avoids invisible version drift and gateway-wide outages); never synthesize a simulated stand-in for missing components. The server knows more than the model about what actually ran — discarding that knowledge makes the stack untrustworthy. Honest caveat: per-call probes are cheap for local imports but network dependencies need TTL-cached capability records.
More from coding & agent
- Codex Can't Route PCB Traces: Devs Still Lack a Workable AI Tool for Hardware's Last Mile — goldvenetianmask · 2026-11-23
- Uncensored 26.4B model runs on a 12GB Mac at 85 tok/s with 8 parallel agents via Locals — SumitGup · 2026-11-23
- spaCy creator Honnibal shares a cheap-LLM ensemble recipe for building NER data with coding agents — honnibal · 2026-11-23
- 398M tokens, 98% cached: how one dev ran 12.5h of agent game dev on 3% of a $200 Claude plan — Yamapama · 2026-11-23
- Harvard paper: higher SWE-Bench scores didn't translate into more merged pull requests — rohanpaul_ai · 2026-11-23
- Solo dev ships AquaPaths puzzle game in 6 months — 9 of 10 commits co-authored by Claude Code — Pleasant_Present_160 · 2026-11-23