AI

See how agents use your MCP server

A new MCP servers page shows the tools agents call, the clients calling them, why calls fail, and each session and user, from the spans your server already sends.

If you run an MCP server, Explore → MCP servers now shows how agents use it. Maple reads the spans your server already emits under the OpenTelemetry MCP conventions. There is no extra SDK or second call to make.

  • The list has every service that answers MCP tool calls, with calls, failure rate, p95, tool count, top client and a volume sparkline.
  • A server’s page charts calls, failures, latency and clients, and breaks them down by tool, client (Claude, Cursor and so on), failure kind and MCP method.
  • A tool’s page groups its errors by type and message, and lists the slowest, failed and recent calls, each linking to its trace.
  • Sessions group a server’s requests by MCP session, in order, linked to the agent session that made them and to the user behind it.

A server that doesn’t send what a view needs gets a checklist of the attributes to add, not an empty table.

Connecting your own agent to Maple is also quicker: the Connect panel in the top bar has a copy-paste command for Claude Code and one-click install links for Cursor and VS Code.

Read more in MCP servers.

See your first trace today.

Add the SDK, point your OTLP exporter at Maple, and traces arrive.

maple.dev: observability on OpenTelemetry