Samply FAQs

Question 1

What is Samply and what does it do?

Accepted Answer

Samply is an MCP server designed for AI agents that enables them to record, load, summarize, and analyze CPU profiles generated by the `samply` tool. It provides a stable interface and structured insights for performance debugging and optimization.

Question 2

What is a typical workflow for an AI agent using Samply?

Accepted Answer

A typical workflow involves calling `samply_record` to capture a profile, then `samply_summarize_profile` to identify overall hotspots. Agents can then use `samply_inspect_thread`, `samply_breakdown_subsystems`, and `samply_focus_functions` for deeper analysis, concluding with `samply_locate_symbols` to pinpoint relevant source code.

Question 3

How does Samply benefit AI agents in performance analysis?

Accepted Answer

Samply offers AI agents a stable API for `samply`, delivers compact, agent-friendly summaries instead of raw profile data, facilitates offline analysis with presymbolication, and provides specialized tools to drill down into performance hotspots and map them directly to source code.

Question 4

What kind of detailed analysis can Samply perform on CPU profiles?

Accepted Answer

Samply can generate summaries of threads and hotspots, inspect specific threads in detail with stack traces, break down native functions by namespace, analyze caller/callee contexts, and precisely map hot native symbols back to their likely local source files for direct code inspection.

Question 5

Does Samply support offline CPU profile analysis?

Accepted Answer

Yes, Samply can load existing `profile.json` files and automatically uses presymbolication sidecar files (e.g., `.syms.json`) to resolve native addresses into function names. This is critical for detailed, AI-driven performance work without requiring live `samply` binary access.

Samply

Samply

Key Features

Use Cases

Key Features

Use Cases