OpenTelemetry MCP Server for Claude

Overview

The OpenTelemetry MCP Server is the official MCP server from Traceloop (makers of OpenLLMetry) that connects Claude to your distributed tracing backend. It supports Jaeger (local/cloud), Grafana Tempo, and Traceloop Cloud as backends. Beyond standard trace search, it includes specialized LLM observability tools: token usage analysis, expensive trace identification, slow LLM operation detection, and AI model discovery. Apache-2.0 licensed, Python 3.11+ required.

Key capabilities

• Trace search — search traces with filters by service, time range, and duration.
• Span search — find individual spans across services.
• Error detection — find traces containing errors or exceptions.
• Service listing — discover all services reporting to the backend.
• LLM usage — aggregate token usage and costs across all LLM calls.
• Model discovery — list all LLM models in use across your system.
• Performance analysis — identify most expensive and slowest LLM operations.

Tools

Configuration

How it compares

OpenTelemetry MCP is the only tracing MCP with dedicated LLM observability tools — token usage, model discovery, and slow AI operation identification — across multiple backends.

Installation

Claude Code (Jaeger local)

claude mcp add opentelemetry \
  -e BACKEND_TYPE=jaeger \
  -e BACKEND_URL=http://localhost:16686 \
  -- uvx opentelemetry-mcp

Claude Code (Grafana Tempo)

claude mcp add opentelemetry \
  -e BACKEND_TYPE=tempo \
  -e BACKEND_URL=https://tempo.example.com \
  -- uvx opentelemetry-mcp

Claude Desktop

{
  "mcpServers": {
    "opentelemetry": {
      "command": "uvx",
      "args": ["opentelemetry-mcp"],
      "env": {
        "BACKEND_TYPE": "jaeger",
        "BACKEND_URL": "http://localhost:16686"
      }
    }
  }
}

Requirements

• OpenTelemetry-compatible backend (Jaeger, Grafana Tempo, or Traceloop).
• Instrumented services sending traces.
• Python 3.11+ with uvx (from uv).
• An MCP client (Claude Code or Claude Desktop).

Security

• Read-only access to your tracing backend.
• Trace data may contain sensitive application data — ensure appropriate access controls on the backend itself.

Source Verification Notes

Verified on 2026-06-18:

• Repository traceloop/opentelemetry-mcp-server (Apache-2.0) on PyPI as opentelemetry-mcp (v0.2.2) documents the uvx opentelemetry-mcp install, BACKEND_TYPE/BACKEND_URL/ BACKEND_API_KEY configuration, support for Jaeger/Tempo/Traceloop backends, all ten tools including the LLM-specific observability tools, Python 3.11+ requirement, and 191 stars.
• Claude Code MCP documentation at code.claude.com/docs/en/mcp describes the stdio connector pattern used above.