Honeycomb MCP

Honeycomb MCP puts an agent in front of Honeycomb observability data, so questions about production behavior can be asked in plain language instead of hand-built query specs. Honeycomb stores wide events, traces and spans; the server exposes the metadata around them, including environments, datasets and column types, plus the query engine itself, so an assistant can pick the right dataset, build a query with calculations, filters, breakdowns and a time range, then read the results back. Beyond raw querying, it reaches the monitoring objects teams already maintain: triggers, SLOs and their burn status, boards and markers. That makes agent-assisted incident work practical. Pull the error rate for a service, break it down by endpoint or customer, follow a slow trace into the spans that caused it, then check whether an SLO is at risk. Because the same agent can also read the repository in an editor, it can cross-reference a hot span with the function that emits it. Honeycomb runs this as a vendor-hosted endpoint with browser-based OAuth, so there are no API keys to distribute; the client picks an environment and an access level at authorization time. An earlier self-hosted implementation was superseded by the hosted service, and access has been gated to the Enterprise tier, so confirm plan eligibility before building a workflow on top of it.