Honeycomb MCP Server

Query and analyze Honeycomb telemetry, traces, SLOs, triggers, and dashboards with natural language, including BubbleUp anomaly investigation.

Observability by Honeycomb OAuth2 active

Connect (Remote Server) GitHub Docs npm: mcp-remote

Overview

Honeycomb MCP connects AI assistants to Honeycomb's observability platform so agents can query traces, metrics, and logs, run BubbleUp investigations, monitor SLOs and triggers, and create boards to document findings. It is offered as a hosted remote MCP server at https://mcp.honeycomb.io/mcp (US) and https://mcp.eu1.honeycomb.io/mcp (EU), with OAuth 2.1 as the recommended authentication method and API key authentication available for unattended agents.

The server enables AI-driven workflows like correlating production behavior with source code, identifying error patterns across distributed traces, retrieving raw rows and individual traces for deep inspection, and proposing instrumentation improvements. Honeycomb's BubbleUp feature, which surfaces statistically anomalous dimensions in a query result set, is exposed as a first-class tool so agents can investigate unusual behavior automatically.

A legacy self-hosted Node.js implementation lives at github.com/honeycombio/honeycomb-mcp but is now deprecated in favor of the hosted service. The hosted server is available across Honeycomb plans through Honeycomb Intelligence enrollment and supports Claude Desktop, Claude Code, Cursor, VS Code, Amazon Q Developer, Windsurf, Goose, and any other MCP-compatible client.

Tools

Tool	Description
`list_datasets`	Enumerate all datasets available in the specified Honeycomb environment.
`get_columns`	Retrieve column metadata (names, types, descriptions) for a dataset.
`run_query`	Execute analytics queries with calculations, filters, breakdowns, and ordering across traces, metrics, and logs.
`analyze_columns`	Run statistical analysis on specific dataset columns to surface value distributions and cardinality.
`bubbleup`	Run a BubbleUp investigation to surface dimensions that differ between anomalous and baseline rows in a query result.
`list_slos`	List all SLOs configured for a dataset.
`get_slo`	Fetch detailed SLO configuration, burn rate, and budget status.
`list_triggers`	List all triggers (alerts) configured for a dataset.
`get_trigger`	Retrieve a trigger's threshold, query, and recipient configuration.
`get_trace_link`	Generate a Honeycomb UI deep link for a specific trace.
`get_instrumentation_help`	Return OpenTelemetry instrumentation guidance for a given language or framework.

Setup Guide

Prerequisites

A Honeycomb account with Honeycomb Intelligence enabled (granted automatically on enrollment)
An MCP-compatible AI client (Claude Desktop, Claude Code, Cursor, VS Code, Amazon Q, etc.)
For API key auth: a team owner generates a key under Account, Team Settings, API Keys with Model Context Protocol and Environments (read, plus write for board creation) scopes

Endpoints

US: https://mcp.honeycomb.io/mcp
EU: https://mcp.eu1.honeycomb.io/mcp

Claude Code CLI

claude mcp add honeycomb --transport http https://mcp.honeycomb.io/mcp

Amazon Q Developer CLI

qchat mcp add --name honeycomb --command npx --args mcp-remote,https://mcp.honeycomb.io/mcp

Generic HTTP MCP host

{
  "mcpServers": {
    "honeycomb": {
      "transport": "HTTP",
      "url": "https://mcp.honeycomb.io/mcp"
    }
  }
}

Clients that require a stdio bridge (e.g. Claude Desktop)

{
  "mcpServers": {
    "honeycomb": {
      "command": "npx",
      "args": ["-y", "mcp-remote", "https://mcp.honeycomb.io/mcp"]
    }
  }
}

API key authentication (unattended agents)

{
  "mcpServers": {
    "honeycomb": {
      "command": "npx",
      "args": [
        "-y",
        "mcp-remote",
        "https://mcp.honeycomb.io/mcp",
        "--header",
        "Authorization: Bearer $HONEYCOMB_API_KEY"
      ],
      "env": {
        "HONEYCOMB_API_KEY": "<KEY_ID>:<SECRET_KEY>"
      }
    }
  }
}

EU customers should swap the URL for https://mcp.eu1.honeycomb.io/mcp.

Use Cases

Investigate a production incident by asking the agent to run a BubbleUp on a spiking error rate to find the offending service, region, or customer.
Audit SLO health across services and surface any objective whose error budget is burning faster than the configured threshold.
Triage paging triggers by listing recently fired triggers, retrieving their underlying queries, and proposing tuning changes.
Correlate trace data with code by retrieving a trace, identifying the slow span, and asking the agent to find the matching function in the repo.
Generate OpenTelemetry instrumentation guidance for a service that is currently missing useful spans or attributes.

Example Prompts

"Run a BubbleUp on the checkout-service dataset for the last 30 minutes where status_code >= 500 and tell me what dimensions are over-represented."
"List every SLO in the prod environment and show me which ones have less than 25% budget remaining."
"Show me the slowest P99 endpoints in the api dataset for the last hour, broken down by route."
"Get the trace for trace_id abc123 in the payments dataset and give me a deep link plus a summary of the slowest span."
"What OpenTelemetry attributes should I add to a Node.js Express service to make Honeycomb queries more useful?"

Pros

Officially built and hosted by Honeycomb, no self-hosting or build step required.
OAuth 2.1 login means no shared API keys for interactive use, with API key fallback for unattended agents.
Exposes BubbleUp, SLOs, and triggers as native tools, not just raw query execution.
Works with every major MCP client (Claude Desktop and Code, Cursor, VS Code, Amazon Q, Windsurf, Goose).

Limitations

Requires a Honeycomb account with Honeycomb Intelligence enabled; not usable without a Honeycomb subscription.
Some clients without native remote MCP support require the mcp-remote stdio bridge, which adds a dependency.
The legacy self-hosted honeycomb-mcp repo is deprecated, so teams that prefer running their own server no longer have a maintained path.

Alternatives

Grafana MCP server (github.com/grafana/mcp-grafana) for querying Loki, Tempo, Prometheus, and Grafana dashboards.
Sentry MCP server (mcp.sentry.dev) for error and performance data.
Datadog community MCP servers for metrics, logs, and APM (no official Datadog-hosted server at this time).