What is the correct approach when an MCP tool call exceeds its per-tool timeout?

Attempt it

Quick reference

TL;DR

MCP defines no timeout, so the host must enforce one: cancel the slow call, feed the model a graceful fallback, and emit a trace span for ops.

Memory aid

Sign in for the mnemonic that makes this stick. See Pro pricing.

Easy to grasp

Imagine you call a friend to look up a fact for you, but they go quiet for ten minutes. You can't wait forever; your own conversation has a clock. So you hang up, tell the person you're helping 'sorry, I couldn't reach my friend in time,' and jot a note that this friend was slow today so you can check on them later. An MCP tool call works the same way. The protocol itself never hangs up for you, so your app has to set its own clock, cancel the call when it runs long, hand the model a clear 'this timed out' message instead of waiting, and write down that the call was slow so the team can investigate.

Key concepts

Latency budget Graceful degradation Observability

Quick scan · 2–4 min

Everything important, quickly.

4 min: latency budget anatomy + three cost components + why MCP has no timeout + the cancel/fallback/trace response + why each distractor fails.

Side-by-side

Strategy on timeout	Effect on this turn	Verdict
Wait for completion	Turn hangs; no protocol timeout exists to save you	Wrong
Cancel, fallback, trace	Budget reclaimed, model informed, outlier logged	Correct
Retry on a replica	Adds another round trip on an over-budget turn	Wrong
Drop silently	Model confabulates around missing data with no signal	Wrong

Real-world examples

Real products, models, and research that use this idea.

Claude Code spawns local stdio MCP servers for hot paths and enforces host-side per-tool deadlines rather than trusting the transport.
Cursor and Zed register remote MCP servers over Streamable HTTP, where network round trips dominate the per-call latency budget.
Production agents wrap MCP tools/call dispatch in OpenTelemetry spans so latency outliers per server surface in dashboards.

Interviewer follow-ups

What an interviewer would ask next. Try answering before peeking at the approach.

QHow would you set per-tool latency budgets across a chain of five MCP calls?

Sign in or go Pro to see the approach for this follow-up.

QWhy does context re-prefill cost grow as an agent runs more tool calls?

Sign in or go Pro to see the approach for this follow-up.

QWhen is retrying a timed-out MCP call actually the right call?

Sign in or go Pro to see the approach for this follow-up.

60-second cheat sheet

The night-before-the-interview bullets. Scan these on the way to the call.

Sign in for 5 unlocks/day or go Pro for unlimited revision sheets. See Pro pricing.

References

Primary sources. Skim if you want the original framing.