Skip to content
NVIDIA CUDA Docs

NVIDIA CUDA DocsMCP Server

by NVIDIA
Added to Onei
1 views
v2026.2.1
Repository Docs
Developer Tools

Summary

NVIDIA's official remote MCP server for CUDA: grounds a coding agent in first-party CUDA documentation and code samples instead of its training data.

Features

  • Searches first-party NVIDIA CUDA documentation and code samples at write time
  • Hosted remote server — nothing to install or keep updated locally
  • One-time NVIDIA Developer account sign-in, reused on later sessions
  • Documented for Claude Code, Codex, Cursor, Antigravity and other MCP clients

Installation

Set up this MCP server in your favorite AI agent — copy a ready-made configuration below.

Any MCP-compatible agent

Most agents (Claude, Cursor, Windsurf, VS Code, and more) read a standard mcpServers configuration.

  1. Open your agent's MCP configuration file.
  2. Merge the snippet below into it.
  3. Restart the agent — the "NVIDIA CUDA Docs" tools will be available.
{
  "mcpServers": {
    "nvidia-cuda-docs": {
      "command": "npx",
      "args": [
        "-y",
        "mcp-remote",
        "https://api.copilot.nsight.ngc.nvidia.com/mcp/cuda-docs"
      ]
    }
  }
}

Description

Part of NVIDIA Nsight AI, this is NVIDIA's official remote MCP server for CUDA documentation. Its purpose is narrow and worth the install: a coding agent writing CUDA from memory produces plausible-looking kernels against APIs that changed two releases ago. This server puts current, first-party NVIDIA documentation and code samples in front of the model at the moment it is writing, so the answer is grounded rather than recalled.

It is a hosted streamable-HTTP endpoint at https://api.copilot.nsight.ngc.nvidia.com/mcp/cuda-docs. On first connection you authenticate with an NVIDIA Developer account; later sessions reuse the credentials, so it is a one-time browser step. NVIDIA documents it working with Claude Code, Codex, Cursor, Antigravity and other MCP-compatible agents.

Because it is a search-and-retrieve server over NVIDIA's own docs, it complements rather than replaces the GPU-profiling and kernel-tuning tooling around it — use it to get the API call right, and a profiler to find out why the kernel is slow.

Related MCP Servers

MCP: Mux MCP Server

by Mux

New

Official MCP server for the Mux video API, built on a code-execution scheme: the agent writes TypeScript against the SDK and runs it in a Deno sandbox.

Developer Tools
1 views

MCP: ripwire

by Red Hat

New

Red Hat's zero-dependency C++23 code-context engine — ranked call graphs and blast-radius analysis, indexing a repo in under half a second with no server and no database.

Developer Tools
2 views

MCP: Graft

by Trail

New

Builds a searchable markdown graph of your repo so coding agents stop re-exploring it on every task — reported 42% fewer tokens and 46% fewer tool calls.

Developer Tools
1 views

MCP: Expo MCP Server

by Expo

New

Expo's official remote MCP server — searches Expo docs, installs compatible SDK packages, triggers and monitors EAS builds, and drives iOS/Android simulators.

Developer Tools
1 views
Browse all MCP servers →