Does NVIDIA CUDA have an MCP server?
NVIDIA CUDA maintains an official MCP server.
Searches CUDA documentation and code samples only; it does not run or profile code.
- Access
- Remote server
- Endpoint
- https://api.copilot.nsight.ngc.nvidia.com/mcp/cuda-docs
- Category
- Developer Tools
A remote server run by NVIDIA CUDA itself — nothing to install. Documented at developer.nvidia.com.
Hosted by the vendor — no install, connect over HTTP with OAuth or a token.
Add it from the terminal, then restart your session:
claude mcp add --transport http nvidiacuda https://api.copilot.nsight.ngc.nvidia.com/mcp/cuda-docsAdd --scope user to make it available in every project.
- Open Settings → Connectors
- Click Add custom connector
- Paste
https://api.copilot.nsight.ngc.nvidia.com/mcp/cuda-docs - Finish the sign-in prompt, then enable it in the chat's tools menu
Add to .cursor/mcp.json in your project, or ~/.cursor/mcp.json for
every project:
{
"mcpServers": {
"nvidiacuda": {
"url": "https://api.copilot.nsight.ngc.nvidia.com/mcp/cuda-docs"
}
}
}A green dot in Settings → MCP means it connected.
Add to .vscode/mcp.json:
{
"servers": {
"nvidiacuda": {
"type": "http",
"url": "https://api.copilot.nsight.ngc.nvidia.com/mcp/cuda-docs"
}
}
}Then pick the server from the tools menu in Copilot's agent mode.
- Enable Developer mode under Settings → Apps & Connectors → Advanced
- Choose Create and paste
https://api.copilot.nsight.ngc.nvidia.com/mcp/cuda-docs - Complete sign-in, then enable the connector in a new chat
Does NVIDIA CUDA have an official MCP server?
Yes. NVIDIA CUDA maintains an official MCP server, hosted at https://api.copilot.nsight.ngc.nvidia.com/mcp/cuda-docs.
How do I connect NVIDIA CUDA to Claude?
Add https://api.copilot.nsight.ngc.nvidia.com/mcp/cuda-docs as a custom connector under Settings → Connectors in Claude Desktop or claude.ai, or run "claude mcp add --transport http nvidia-cuda https://api.copilot.nsight.ngc.nvidia.com/mcp/cuda-docs" in Claude Code.
Do I need to install anything to use the NVIDIA CUDA MCP server?
No. It is a remote server hosted by NVIDIA CUDA, so your MCP client connects to it over HTTP. You authenticate with the vendor rather than installing a package.
How is this verdict verified?
Published to the MCP Registry under NVIDIA's verified namespace com.nvidia; the CUDA MCP Server and its endpoint are documented on developer.nvidia.com/nsight-ai, and the endpoint answers MCP initialize with 401 (auth required). GitHub metadata is refreshed every six hours.
Published to the MCP Registry under NVIDIA's verified namespace com.nvidia; the CUDA MCP Server and its endpoint are documented on developer.nvidia.com/nsight-ai, and the endpoint answers MCP initialize with 401 (auth required).
- github data refreshed 2026-09-29, then every 6 hours
- vendor docs verified on 2026-09-29
Spotted something out of date? Tell us — corrections ship same-day.