Uniprof

🚀 uniprof

uniprof simplifies CPU profiling for humans and AI agents. It allows you to profile any application without code changes or added dependencies.

🚀 Quick Start

Profile and analyze in one step

# Profile most languages and get immediate analysis
uniprof python app.py
uniprof node server.js
uniprof ruby script.rb
uniprof java -jar myapp.jar
uniprof dotnet MyApp.dll
uniprof ./my-native-app

# Profile and visualize flamegraph in browser
uniprof --visualize python app.py

Save profiles to analyze and visualize later

# 1. Check environment (optional)
uniprof bootstrap

# 2. Record a profile
uniprof record -o profile.json -- python app.py
uniprof record -o profile.json -- node server.js
uniprof record -o profile.json -- ruby script.rb
uniprof record -o profile.json -- php script.php
uniprof record -o profile.json -- java -jar myapp.jar
uniprof record -o profile.json -- ./gradlew run
uniprof record -o profile.json -- ./mvnw spring-boot:run
uniprof record -o profile.json -- dotnet MyApp.dll
uniprof record -o profile.json -- elixir script.exs
uniprof record -o profile.json -- mix run
uniprof record -o profile.json -- ./my-native-app
uniprof record -o profile.json -- /Applications/MyApp.app

# 3. Analyze profile data to find hotspots
uniprof analyze profile.json

# 4. Visualize flamegraph in the browser
uniprof visualize profile.json

For detailed command line options documentation see docs/cli.md.

Host vs. Container Modes

Use the --mode option to control how profiling runs:

• auto (default): Prefer container mode when Docker is available; otherwise use host.
  • Language runtimes (Python/Node.js/Ruby/PHP/BEAM/JVM/.NET) default to container for zero-setup.
  • Native on macOS with Mach-O binaries uses host (Instruments). ELF binaries on macOS are supported in container mode.
  • Native on Linux defaults to container; host can be used if you prefer your local perf setup.
• host: Force host-installed profilers.
• container: Force Docker containers (not supported for macOS Instruments/Mach-O binaries).

# Auto mode (default)
uniprof record -o profile.json -- python script.py

# Force host profilers
uniprof record --mode host -o profile.json -- python script.py

# Force container mode
uniprof record --mode container -o profile.json -- ./my-linux-app

✨ Features

uniprof implements a common interface over multiple profilers that specialize in different platforms and runtimes. It automatically detects which profiler to use based on the command being executed, runs the profiler, transforms the varying output formats into a single format, and runs statistical analysis on the data to identify hotspots.

📦 Installation

npm install -g uniprof

MCP Server

# Install uniprof MCP server automatically
uniprof mcp install claudecode
uniprof mcp install cursor
uniprof mcp install vscode

Supported clients for auto-installation: amp, claudecode, codex, cursor, gemini, vscode, zed

If auto-installation is not supported for your client, add an MCP server using the stdio transport with the command npx -y uniprof mcp run.

For detailed MCP documentation see docs/mcp.md.

📚 Documentation

• CLI Reference
• Advanced Usage
• Platform Guides
• MCP Server

🔧 Technical Details

Supported Platforms

* Linux only for host mode

System Requirements

tl;dr: macOS or Linux with Docker installed

uniprof is designed to run on macOS and Linux but has primarily been developed & tested on macOS. If you find any bugs when running on Linux, please report them. Running on a Windows host is currenly not supported, but you can run on Linux installed on Windows via WSL2. Note that when using uniprof via WSL2, profiling native Windows executables is not supported, but you can still use the profilers for the higher level language runtimes like Python and Node.js.

Profiling tools are often non-trivial to set up correctly and can require elevated privileges. To simplify set up and provide better isolation for the profiled code, uniprof defaults to using Docker containers for each runtime that are pre-configured to run specific profiling tools. uniprof mounts your workspace in the container and executes the program with the profiler attached. uniprof also supports profiling outside the container if the host system has the profiling tools installed by running with the --mode host flag. The only case where containerized execution is not supported is when profiling a native Mach-O binary on macOS, since Apple Instruments cannot run inside a container.

Compiling with Debug Information

Native profiling requires debug information for meaningful results. Debug symbols enable the profiler to map memory addresses to function names, providing readable output instead of just hexadecimal addresses. Additionally, frame pointers improve call stack accuracy, especially for optimized code.

Credits

• py-spy (MIT) is used for Python profiling. Included in the uniprof-python image.
• 0x (MIT) is used for Node.js profiling. Included in the uniprof-nodejs image.
• rbspy (MIT) is used for Ruby profiling. Included in the uniprof-ruby image.
• Excimer (Apache 2.0) is used for PHP profiling. Included in the uniprof-php image.
• async-profiler (Apache 2.0) is used for JVM profiling. Included in the uniprof-jvm image.
• dotnet-trace (MIT) is used for .NET profiling. Included in the uniprof-dotnet image.
• perf (GPL) is used for native profiling on Linux. Included in the uniprof-native and uniprof-beam images.
• speedscope (MIT) is bundled with uniprof and used for flamegraph visualization

📄 License

All uniprof source code in this repository is licensed under the MIT License. See LICENSE.

uniprof downloads and runs Docker images that bundle third-party software. The Dockerfiles used to build these images can be found in containers/. These images and their contents are not covered by the MIT license and are governed by their own licenses. uniprof does not grant rights to those components and does not relicense them.