MCP tool failures hide in HTTP 200, agent retries 6 times billed with no error traces
Thirumalaiboobathi · reddit · 2026-09-09
A production failure pattern in MCP setups: failed tools return HTTP 200 with isError: true inside the JSON-RPC payload, so generic OTel instrumentation marks spans OK and dashboards stay green while every call fails.
Why it burns money: the model sees the error as normal tool output, rephrases and retries; each retry carries the grown context, so six billed calls produce nothing and only surface on the monthly bill.
The fix (opentel-mcp, open source):
- Mark spans ERROR when isError is set despite transport success
- Fingerprint failures (normalize, hash) so the same root cause groups across varying messages
- Detect retry loops: same fingerprint N times becomes one event with loop length and tokens burned
- Per-tool-call cost attribution, sortable by spend
- Flag silent inputSchema drift
- Works on both MCP SDK generations including the stateless spec
More from coding & agent
- Open-source Switch Buddy puts Claude Code, Codex and other agents in shared rooms — anthara_ai · 2026-11-19
- Stanford CS329Z Lecture 5 compares DSPy vs. LangGraph agent framework approaches — stanfordnlp · 2026-11-19
- Mistral Large 4 coding test: 2 hours of agentic work burned just €0.82 of quota — qtnx_ · 2026-11-19
- Railway breach investigator: LLM attacker's infrastructure still targeting other AI companies — cyb3rops · 2026-11-19
- Gradio and HF launch hackathon: reproduce NeurIPS 2026 papers with AI agents — Gradio · 2026-11-19
- NVIDIA open-sources SkillEvaluator to vet AI agent skills before deployment — anthara_ai · 2026-11-19