Honeycomb MCP Server (Deprecated) — Migrate to Hosted

Overview

This repository provided a self‑hosted Model Context Protocol (MCP) server that exposed Honeycomb observability data to LLMs and other MCP clients. The server translated MCP tool calls into Honeycomb API requests so clients like Claude, Cursor, and others could query datasets, inspect schemas, run analytics, and surface SLO/trigger metadata.

Important: this project is deprecated. Honeycomb now offers a hosted MCP solution with guided migration and support. If you are still running a self‑hosted instance, plan to migrate — see Honeycomb MCP Documentation for migration steps and timelines: https://docs.honeycomb.io/integrations/mcp/

Features

• Expose Honeycomb datasets via MCP-compatible tool endpoints
• Run analytics queries: counts, percentiles (P95), averages, breakdowns, filters, and time windows
• Inspect dataset schema and column metadata
• List and retrieve SLOs and Triggers (Enterprise customers)
• Generate deep links to traces in the Honeycomb UI
• Column analysis helpers (basic statistics, histograms)
• TTL-based caching for non-query API calls to reduce API usage and improve responsiveness
• Works with multiple MCP clients (Claude Desktop/Code, Cursor, Windsurf, Goose, etc.)

Requirements

• Node.js 18 or newer
• Honeycomb API key(s) with broad permissions:
  • Query access (analytics)
  • Read access for SLOs and Triggers
  • Environment-level access for dataset metadata
• Intended for Honeycomb Enterprise customers only

Installation / Configuration

Build and run the server locally:

# install dependencies (pnpm is used in the original project)
pnpm install

# produce build artifacts
pnpm run build

# the build output is placed under ./build

Embed the server in your MCP orchestration by launching the generated Node process and supplying Honeycomb credentials via the MCP config env block. Example MCP server config snippet:

{
  "mcpServers": {
    "honeycomb": {
      "command": "node",
      "args": ["/full/path/to/honeycomb-mcp/build/index.mjs"],
      "env": {
        "HONEYCOMB_API_KEY": "your_api_key_here"
      }
    }
  }
}

Multiple environments (staging/production) can be provided as separate env variables:

{
  "mcpServers": {
    "honeycomb": {
      "command": "node",
      "args": ["/full/path/to/honeycomb-mcp/build/index.mjs"],
      "env": {
        "HONEYCOMB_ENV_PROD_API_KEY": "prod_key",
        "HONEYCOMB_ENV_STAGING_API_KEY": "staging_key"
      }
    }
  }
}

EU customers must override the API endpoint (default is https://api.honeycomb.io):

# example: EU API endpoint
export HONEYCOMB_API_ENDPOINT=https://api.eu1.honeycomb.io/

Caching configuration

The server caches non-query responses; configure via environment variables:

Set these in the same env block used to launch the MCP process.

Available Tools / Resources

The server exposed several MCP tools (JSON input shown):

• list_datasets
  • Input: { “environment”: “production” }
• get_columns
  • Input: { “environment”: “production”, “dataset”: “api-requests” }
• run_query
  • Input example:

    {
      "environment": "production",
      "dataset": "api-requests",
      "calculations": [
        { "op": "COUNT" },
        { "op": "P95", "column": "duration_ms" }
      ],
      "breakdowns": ["service.name"],
      "time_range": 3600
    }
    

• analyze_columns
  • Runs statistical queries for selected columns and returns metrics/histograms
• list_slos / get_slo
  • Input: environment, dataset, optional sloId
• list_triggers / get_trigger
  • Input: environment, dataset, optional triggerId
• get_trace_link
  • Generates a Honeycomb UI deep link for a trace
• get_instrumentation_help
  • Provides guidance for OpenTelemetry instrumentation

Resource URI format used by the server: honeycomb://{environment}/{dataset} (e.g., honeycomb://production/api-requests)

Use Cases

• Assist an LLM to explore dataset schema before composing queries: call get_columns, then run_query.
• Let a code-assistant examine recent SLO status and link to related triggers for incident triage.
• Drive automated dashboards: list_datasets → run_query for aggregated metrics → render in external tooling.
• Debug traces via conversational tools: use get_trace_link to open a focused trace in the Honeycomb UI from an LLM session.

Migration (Deprecated server → Hosted)

Because this self-hosted MCP server is deprecated, follow these high-level steps to migrate to Honeycomb’s hosted MCP:

1. Review Honeycomb MCP Documentation for the hosted offering and migration timelines: https://docs.honeycomb.io/integrations/mcp/
2. Inventory current MCP usage: environments, datasets, API keys, and client integrations.
3. Map environment API keys and update permissions if needed for the hosted integration.
4. Switch your MCP clients from the local server to the hosted endpoint per the Honeycomb docs.
5. Decommission local MCP processes once traffic is confirmed on the hosted service.

For detailed guidance, step‑by‑step instructions, and deadlines, consult the official Honeycomb MCP documentation linked above. For troubleshooting migration issues, contact Honeycomb support.