Honeycomb Canvas Connectors GA: a dozen tools for agent context

Honeycomb’s blog (2026-09-29, Dan Juengst) made Canvas Connectors generally available—framed as “12 connectors and growing” / a dozen, available to all Honeycomb customers under Settings > Connectors—and shipped Anomaly Detection to GA in the same post (Honeycomb blog).

Canvas Connectors (GA)

Connectors let the Canvas agent reach context outside Honeycomb telemetry: code, incidents, runbooks, tickets, and similar sources. Honeycomb says connectors are built on MCP, the same open protocol as Honeycomb’s own MCP server, which is why the roster “will keep growing” (Honeycomb blog).

Named examples from this GA post (Honeycomb does not list all twelve): GitHub, PagerDuty, Jira, Amplitude, and Confluence. Count framing here stays at “12 and growing.”

As Honeycomb states: reads are allowed by default once a tool is connected; writes ask for permission each time, with a human approving the action—a vendor claim about the permission UX (Honeycomb blog).

Anomaly Detection (GA)

Anomaly Detection is generally available. Honeycomb says it learns a statistical baseline per service for error rate and presence, with nothing to configure manually, auto-onboards services with enough steady traffic, and is available for every team. Request rate, latency, and seasonality-aware detection are framed as coming soon—not this GA (Honeycomb blog).

AI Ecosystem = early access (not GA)

AI Ecosystem (fleet-wide AI agent performance and cost views) and LLM cost tracking are early access only—not generally available in this announcement. Honeycomb says cost figures are estimates from model/token spans and public list prices, useful to see what drives spend, not to reconcile a provider invoice (Honeycomb blog).

Soft siblings

Also in the same post (not the lead): Canvas extensions to create/update triggers, SLOs, and boards (edits need human approval); onboard instrumentation via Honeycomb MCP from Claude Code, Cursor, Codex, and others; and a donation of adaptive tail-sampling algorithms to the OpenTelemetry Collector, with Refinery still described as the more complete sampling proxy (Honeycomb blog).

Who should care

Honeycomb customers who want Canvas agents to pull stack context beyond telemetry should start at Settings > Connectors via the GA post. Lead with Connectors and Anomaly Detection GA; treat AI Ecosystem and LLM cost tracking as early access.