wtf is opentel-mcp?

opentel-mcp is a JavaScript library that adds observability to servers built with the Model Context Protocol, or MCP, a standard that lets AI agents like Claude call external tools. When an AI agent calls several tools in a row, it can be hard to know which tool was slow or which one failed silently. opentel-mcp solves that by wrapping an MCP server so that every tool call automatically generates a trace using OpenTelemetry, an open industry standard for tracking what a piece of software is doing. With opentel-mcp installed, each tool call becomes a labeled entry you can view in observability tools such as Jaeger, Grafana Tempo, Honeycomb, Datadog, or SigNoz. Instead of only knowing that an entire agent turn took two and a half seconds, you can see exactly which individual tool call caused the delay. The library also catches a tricky kind of failure: an MCP tool can technically succeed while still reporting an error inside its response. Most tools would record that as a success, but opentel-mcp marks it correctly as an error, so your error rate reflects what actually happened. Beyond traces, the library also emits a small set of metrics, such as how many times each tool was called, how many errored, and how long each call took, using OpenTelemetry's Metrics API. It does not include or require its own tracing backend, it only produces standard data that plugs into whatever monitoring stack you already use. Setup is a short code change: install the package with npm, then call a single instrumentation function before registering your MCP server's tools. It supports both the low level and high level MCP server APIs. The project is still in early development, at version 0.x, and the underlying tracing conventions it follows are not yet finalized, so some details may change before a stable 1.0 release. This is a small, focused developer tool with 2 stars, useful for anyone building or debugging MCP servers who wants to understand what their AI agent's tool calls are actually doing.