NVIDIA Nsight AI
Bring NVIDIA knowledge and tools into the AI-assisted workflows you use to build, profile, and improve your code, with options that fit how and where you work.
Choose Your Nsight AI Workflow
CUDA MCP Server
Connect your preferred AI coding agent to current, first-party NVIDIA CUDA documentation and code examples through the NVIDIA-hosted CUDA MCP Server.
Nsight AI Blueprint
Deploy an open-source, self-hosted CUDA AI backend in your own environment and connect compatible IDE clients.
Nsight Compute Integration
Use Nsight AI in Nsight Compute for interactive, AI-assisted guidance as you analyze CUDA kernel performance.
Explore the IntegrationSee Nsight AI in Action
See how the Nsight AI Blueprint supports a self-hosted NVIDIA CUDA™ AI workflow and how Nsight AI in Nsight Compute provides interactive performance guidance.
Watch the Nsight AI Blueprint Demo
This video shows the open source Nsight Copilot Blueprint, a self-hosted CUDA AI backend built for NVIDIA GPU-accelerated systems.*
Watch Nsight Copilot in Nsight Compute
This video demonstrates Nsight Copilot integration in Nsight Compute, providing interactive guidance on uncoalesced memory accesses.
Get Started
Connect the CUDA MCP Server
Choose your agent and connect. On first connection, sign in with your NVIDIA Developer account. Your client will reuse that authentication afterward.
Run this command from you terminal
claude mcp add --scope user --transport http nvidia-cuda-docs https://api.copilot.nsight.ngc.nvidia.com/mcp/cuda-docs
Run this command from you terminal
codex mcp add nvidia-cuda-docs --url https://api.copilot.nsight.ngc.nvidia.com/mcp/cuda-docs
Add to your custom MCP configuration
{
"mcpServers": {
"nvidia-cuda-docs": {
"url": "https://api.copilot.nsight.ngc.nvidia.com/mcp/cuda-docs"
}
}
}
Add to your custom MCP configuration
{
"mcpServers": {
"nvidia-cuda-docs": {
"serverUrl": "https://api.copilot.nsight.ngc.nvidia.com/mcp/cuda-docs"
}
}
}
Connect to the CUDA MCP Server
https://api.copilot.nsight.ngc.nvidia.com/mcp/cuda-docs
Deploy the Blueprint
Run an open source, self-hosted CUDA AI backend in your own environment, then connect a compatible IDE client.*
1. Review the Requirements
Review the current hardware and software requirements in the GitHub deployment guide before you begin.
2. Deploy the Nsight AI Blueprint
Follow the GitHub deployment guide to deploy the Blueprint with Docker Compose.
3. Connect Your IDE
Connect a compatible IDE client to your running Blueprint deployment.
NVIDIA Nsight AI FAQ
Get answers to common questions about NVIDIA Nsight AI, including the CUDA MCP Server, self-hosted deployment, Nsight Compute integration, and system requirements.
The Nsight Copilot extension for Visual Studio Code is being deprecated. To bring current NVIDIA CUDA knowledge into your preferred AI coding agent, connect to the NVIDIA-hosted CUDA MCP Server. For a self-hosted CUDA AI workflow, deploy the open-source Nsight Copilot Blueprint and connect a compatible client. This change applies only to the Visual Studio Code extension; Nsight Copilot in Nsight Compute is not affected.
Explore the updated capabilities on the NVIDIA Nsight AI product page.
The CUDA MCP Server provides a standardized interface that connects your preferred AI coding agent to real-time, first-party NVIDIA CUDA documentation and code examples, ensuring the model provides accurate and context-aware answers.
Review the documentation on how to connect your agent to the CUDA MCP Server.
The Nsight Copilot Blueprint is an open-source, self-hosted CUDA AI backend optimized for NVIDIA GPU-accelerated systems, allowing developers to maintain full control over their development environment and data privacy while utilizing advanced AI assistance.
Download and deploy the Nsight Copilot Blueprint from the GitHub repository.
You can use the Nsight Copilot integration within Nsight Compute to receive interactive guidance specifically focused on memory analysis, such as identifying and resolving uncoalesced memory accesses in your CUDA kernels.
Get started with the Nsight Compute integration tools.
Yes, on your first connection to the CUDA MCP Server, you must authenticate using your NVIDIA Developer account; once authenticated, your client will securely manage and reuse these credentials for subsequent sessions.
Join the NVIDIA Developer Program to manage your account credentials.
The Blueprint requires an NVIDIA GPU-accelerated system running Ubuntu 22.04 or later, Docker Compose v2, and the NVIDIA Container Toolkit, along with a minimum of 200 GB of free disk space for the deployment.
Confirm your current system requirements via the GitHub deployment documentation.
The hosted CUDA MCP Server provides access to NVIDIA-curated documentation; however, users handling highly sensitive or proprietary code should use the self-hosted Nsight Copilot Blueprint to ensure data remains strictly on-premises.
Learn about self-hosted options on the Nsight Copilot Blueprint GitHub.
No, Nsight AI complements traditional debugging by providing real-time, AI-powered insights and analysis, but it should be used in conjunction with standard debugging tools for comprehensive application validation.
Learn more about the tools overview for Nsight Compute.
MCP is the standardized communication protocol used by the CUDA MCP Server, allowing various coding assistants to securely interface with NVIDIA's knowledge base and documentation.
Understand the protocol usage in the Nsight Copilot GitHub documentation.
More Nsight Resources
Get started with Nsight AI
Nsight Copilot BlueprintCUDA MCP ServerNsight Compute IntegrationVS Code Extension
Support
To provide feedback, request additional features, or report support issues, please use the Developer Forums.
*Hardware compatibility depends on the requirements of the included NIM™ models. See the GitHub guide for current system requirements.