p:cuda-mcp

Navigate CUDA symbols and diagnose issues in .cu/.cuh files via clangd.

1|Updated Jan 7, 2026
One-click install
npx skills add https://github.com/pontscho/prompt-heaven --skill p-cuda-mcp
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: p:cuda-mcp
Source: https://github.com/pontscho/prompt-heaven/tree/main/ClaudeCode/skills/p%3Acuda-mcp
Command: npx skills add https://github.com/pontscho/prompt-heaven --skill p-cuda-mcp

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve?

Navigating CUDA source and diagnosing symbol definitions with reliable accuracy using a single CUDA-aware tool, instead of deprecated text-search methods.

Core Features & Use Cases

  • Compiler-accurate CUDA code intelligence via clangd with automatic CUDA SDK/arch discovery
  • Find definitions, references, hover, diagnostics, and workspace symbols in CUDA (.cu/.cuh) files
  • Use cases: kernel development, CUDA SDK debugging, large translation unit navigation

Quick Start

Use cuda_call to initialize and perform a symbol query for a CUDA kernel, without ever resorting to grep or text search.

Frequently Asked Questions about p:cuda-mcp

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I navigate CUDA kernel symbols and find definitions in .cu files?

You can navigate CUDA kernel symbols by applying compiler-aware indexing via clangd, which provides definition references, hover, diagnostics, and workspace symbol searches across .cu and .cuh files without relying on text search.

Does clangd work with CUDA source files for code navigation and diagnostics?

Yes, clangd works with CUDA source files by utilizing CUDA SDK awareness and auto-discovery of CUDA_PATH, nvcc, and arch to deliver accurate code navigation and diagnostics for device code and kernels.

What is the best way to find references in large CUDA translation units without grep?

The best way to find references in large CUDA translation units is using a single CUDA-aware code intelligence tool that implements compiler-accurate indexing, replacing deprecated text-search methods like grep.

How do I set up CUDA code navigation with automatic SDK and nvcc discovery?

Setting up CUDA code navigation involves leveraging a tool that automatically discovers your CUDA_PATH, nvcc, and target architecture, allowing you to initialize and perform symbol queries on CUDA kernels immediately.

Why do I get inaccurate diagnostics when navigating CUDA device code?

Inaccurate diagnostics for CUDA device code often occur when using deprecated text-search methods instead of compiler-aware indexing tools that understand the CUDA SDK, nvcc, and architecture-specific definitions.

Can I perform workspace symbol searches across multiple .cuh header files?

Yes, you can perform workspace symbol searches across multiple .cuh header files and .cu source files to locate kernel definitions and device code symbols using a unified CUDA-aware navigation tool.