Bring Lunary docs into your editor. Trace the tools you build.

At a glance

Three different jobs often called an MCP integration

A documentation MCP server helps an assistant find implementation instructions. A data MCP server lets an assistant work with application data using the tools and permissions that server exposes. MCP tracing records what happens when an application uses tools. These are separate capabilities. Lunary's published public server is for the first job; connecting it does not automatically enable the other two.

Connect the official documentation server

For Cursor, the official Lunary documentation shows the configuration below in your MCP settings. Other clients have different configuration formats, so use the linked setup guide for yours. After connecting, ask the assistant to find the Lunary ingestion documentation and describe the required fields with a source link. Verify that it uses the server and returns the intended documentation rather than answering only from memory.

{
  "mcpServers": {
    "lunaryDocs": {
      "url": "https://docs.lunary.ai/mcp"
    }
  }
}

What the connection can and cannot do

The Docs MCP server supplies read-only search and page content from Lunary developer documentation. The documentation explicitly says it does not call the Lunary API for you. Do not give it a project API key expecting production trace access, or describe a successful documentation lookup as proof of application instrumentation. If you need programmatic project access, evaluate the documented API and your authorization requirements as a separate integration.

Trace actual MCP tool execution

For a tool-using application, capture the parent task, the model decision, and the tool execution as distinct operations. Lunary's manual instrumentation documents tool runs and parentRunId, while its OpenTelemetry integration supports an existing tracing path. Record the tool name, safe input summary, execution result, duration, and error where appropriate. A tool-call argument emitted by the model does not prove that the tool ran successfully.

Verify a useful end-to-end task

Use a small tool workflow with a successful call and a deliberate safe failure. Check that the actual execution appears beneath the right parent task and that the final user-visible result agrees with the trace. Include an asynchronous or cancelled task if your application supports it. Review payload handling before widening the rollout, and verify that tokens or credentials are not embedded in the fields you chose to log.

• Documentation acceptance: the editor retrieves a relevant sourced Lunary page.
• Runtime acceptance: a real tool execution and its outcome are visible.
• Relationship acceptance: model and tool operations belong to the intended task.
• Failure acceptance: an error is preserved without inventing a successful result.

If you need an assistant to query stored telemetry

Be precise about that requirement when comparing products. Langfuse and Helicone both document data MCP servers. Lunary's public Docs MCP server is a different offering. Choose based on the tools, authentication, and access scope you need, and treat any custom API connection as its own implementation rather than an implied feature of this documentation server.

Common questions