Honeycomb MCP Server – Observability AI Access

Overview

Honeycomb MCP is Honeycomb's hosted Model Context Protocol server for connecting AI agents to Honeycomb observability data. It is designed for engineering and operations workflows where an agent needs live context from Honeycomb rather than relying on static documentation or pasted screenshots. Once connected, an agent can inspect telemetry, query behavior over time, and use Honeycomb-specific investigation tools from an IDE, CLI, or assistant interface.

What the MCP server enables

The server exposes Honeycomb tools for workspace discovery, query and analysis, traces and spans, service maps, semantic conventions, Boards, Triggers, Service Level Objectives, notification recipients, Canvas investigations, and AI conversation analysis. Typical read operations include discovering environments and datasets, running queries, retrieving trace waterfalls, listing spans, checking SLOs, reviewing Triggers, and analyzing prior AI-agent conversations captured with OpenTelemetry attributes. Write-capable tools can create or update Boards, Triggers, SLOs, notification recipients, and Canvas investigations when the connection has the required write scope.

When to use it

Use Honeycomb MCP when an AI agent should help diagnose latency spikes, error bursts, noisy alerts, slow services, or unusual traffic patterns using actual Honeycomb data. It is also useful for translating existing dashboards or alerts into Honeycomb queries, exploring schemas and semantic conventions, generating investigation notes on Boards, and helping teams preserve incident findings in a shared Honeycomb workspace.

Connection and authentication

Honeycomb provides hosted HTTP MCP endpoints for the US and EU regions. OAuth 2.1 is the recommended authentication method for interactive agents because it avoids sharing API keys. For headless or unattended agents that cannot complete OAuth, Honeycomb supports API key authentication using a Management API key with Model Context Protocol and Environments scopes. API-key access should use the narrowest required scopes; write access is needed only for mutating tools.

Key considerations

A Honeycomb team must have Honeycomb Intelligence enabled before using MCP. Most tools are read-only by default, while write tools require explicit mcp:write scope. Some tools depend on plan tier or feature availability, such as Enterprise service map access or anomaly detection. Honeycomb enforces per-tool rate limits, with lower limits for expensive or mutating operations, so agents should avoid unnecessary polling and repeated broad queries.