Skip to main content
If your application exposes an MCP server, Monocle provides a dedicated dashboard for tracking how AI tools interact with it. See which tools are called, by whom, and how they perform.
This page is about monitoring your MCP server. If you are looking for how to connect AI tools (Claude Code, Cursor, etc.) to Monocle’s MCP server, see the Monocle MCP Server page.

​
What Monocle tracks

  • Overview stats: total tool calls, average and P95 latency, error rate, unique users and clients
  • Calls and duration charts: time-series showing success/error distribution and latency trends
  • Clients table: which MCP clients are making calls, with volume and timing data
  • Users table: which users are using the MCP server
  • Tools table: breakdown by tool with call count, latency percentiles, and error rate. Click any tool to see its detailed metrics and recent calls with links to full traces.

​
Span attributes

Since there is no official OpenTelemetry semconv for MCP yet, attributes use the mcp.* namespace. All MCP instrumentations must emit these attributes for the dashboard to work correctly.

​
Core attributes

​
Identity attributes

​
Target attributes

​
Monocle attributes

​
Error attributes

These follow the RPC Semantic Conventions:

​
Setup

@modelcontextprotocol/sdk

Install and configure @monocle.sh/instrumentation-mcp