Conversational Honeycomb Trace Explorer via MCP

Gives engineers a conversational analyst for latency. Instead of a single-shot query, the agent reasons across turns — running a Honeycomb query through a custom MCP tool, reading the p95 breakdown, then drilling into the slowest service on its own to find the offending span. It explains each step so the human follows the chain of reasoning.

When the question isn't "what's the p95" but "why is it slow" — the kind of open-ended investigation that needs several follow-up queries. Use it for deeper debugging sessions where one query is never enough.

1. 1An engineer starts a chat thread with the bot describing the latency symptom.
2. 2The agent calls a custom MCP server exposing Honeycomb query tools to fetch an initial p95 breakdown.
3. 3The agent reasons over results and decides whether to drill into a slower service or span.
4. 4It issues follow-up Honeycomb queries through MCP until it isolates the dominant contributor.
5. 5The agent replies in Slack with the root-cause span, the supporting p95 numbers, and the query trail it followed.