Creativity · MCP — server

MCP OpenTelemetry Server

OpenTelemetry (OTel) is the CNCF standard for traces, metrics, and logs. The MCP OpenTelemetry Server lets an LLM explore trace data from any OTLP-compatible backend (Jaeger, Tempo, Honeycomb, etc.) — search spans, compute latency histograms, and follow distributed request paths through tool calls.

MCP facts

Kind: server
Ecosystem: anthropic-mcp
Language: Go
Transports: stdio

Capabilities

Tools: search_spans, get_trace, service_graph, metric_query
Backends: Jaeger, Grafana Tempo, Honeycomb, New Relic (via OTLP)
Supports bearer-token auth to upstream backends

Install

go install github.com/community/mcp-otel@latest

Configuration

{
  "mcpServers": {
    "otel": {
      "command": "mcp-otel",
      "env": {
        "OTEL_BACKEND": "tempo",
        "OTEL_ENDPOINT": "https://tempo.example.com",
        "OTEL_TOKEN": "<token>"
      }
    }
  }
}

Frequently asked questions

Why use this instead of a vendor-specific server?

Portability. If you move from Tempo to Honeycomb or Jaeger, the same MCP tool surface keeps working — only the backend config changes.

Does it scale to production volumes?

The MCP server is a thin proxy. Scaling limits are the upstream backend's. The server caps result sizes to protect token budgets.

Can it correlate traces with logs?

If the backend supports trace-to-logs correlation, yes — via the get_trace tool which includes log-link annotations.

Sources

OpenTelemetry Documentation — accessed 2026-04-20
Model Context Protocol — accessed 2026-04-20