NVIDIA CUDA Docs, verified
MCP server by NVIDIA · Verified
Search current first-party NVIDIA CUDA documentation and code samples from your AI coding agent.
Connect to NVIDIA CUDA Docs
NVIDIA CUDA Docs needs sign-in. Connect with its own URL:
Your client opens NVIDIA’s sign-in the first time you use it. The token stays between your client and NVIDIA.
Share this server
Opens this page, with the URL and setup steps.
mcp.tc/i/nvidia-cuda-docsAbout
Connects an AI coding agent to NVIDIA's first-party CUDA documentation and code examples, so answers about CUDA development come from current sources instead of model memory. It is part of NVIDIA's Nsight AI offering.
The server is hosted by NVIDIA and uses the streamable HTTP transport. On first connection you sign in with an NVIDIA Developer account through OAuth, and the client reuses that sign-in afterward. Nothing needs to be installed locally.
What you can do
- Search NVIDIA CUDA documentation
- Find CUDA code samples and examples
- Ground CUDA answers in current first-party sources
- Use from agents such as Claude Code, Codex and Cursor
Tools
NVIDIA CUDA Docs lists its tools only after you sign in, so we can’t show them here. Your client lists them once you’re connected.
Example prompts
How do I avoid uncoalesced global memory accesses in my CUDA kernel?
Find an NVIDIA code sample for using shared memory in a matrix multiply.
What does the CUDA docs say about cudaMallocAsync?
Show the recommended way to launch a cooperative groups kernel.
Set up
Every client connects to api.copilot.nsight.ngc.nvidia.com, and NVIDIA asks you to sign in the first time. Pick yours; the page remembers your choice.
Claude Code
- Run this in a terminal, in your project folder:
claude mcp add --transport http nvidia-cuda-docs https://api.copilot.nsight.ngc.nvidia.com/mcp/cuda-docs- Start Claude Code, type
/mcp, pick nvidia-cuda-docs and choose Authenticate. A browser window opens for the NVIDIA sign-in.
Add --scope user to make it available in every project, not just this one.
Claude Desktop
- Open Customize → Connectors, click + Add, then Add custom connector.
- Name it NVIDIA CUDA Docs and paste this URL:
https://api.copilot.nsight.ngc.nvidia.com/mcp/cuda-docs- Click Add, then Connect, and sign in when NVIDIA asks.
Claude Desktop’s JSON config file only starts local servers. Remote servers go through Connectors, and connectors you add on claude.ai show up here too.
claude.ai
- Open the connector form on claude.ai. This button fills in the name and URL for you:
Add to claude.ai (opens in a new tab)
- Check that the URL reads
https://api.copilot.nsight.ngc.nvidia.com/mcp/cuda-docsand click Add. - Click Connect and sign in when NVIDIA asks.
Free plans allow one custom connector. On Team and Enterprise plans an owner adds it under Organization settings → Connectors.
ChatGPT
- On chatgpt.com, go to Plugins, select +, then Add custom MCP server.
- Enter a name for the server.
- Under Connection, paste
https://api.copilot.nsight.ngc.nvidia.com/mcp/cuda-docsas the server URL, then for authentication choose OAuth. ChatGPT sends you to NVIDIA to sign in. - Check I understand and want to continue, then select Create as a plugin.
Whether you can add a custom MCP server depends on your ChatGPT plan and your workspace permissions (in Business workspaces, admins and owners can).
Cursor
Add to Cursor (opens in a new tab)
Or add it by hand to ~/.cursor/mcp.json (all projects) or .cursor/mcp.json (this project):
{
"mcpServers": {
"nvidia-cuda-docs": {
"url": "https://api.copilot.nsight.ngc.nvidia.com/mcp/cuda-docs"
}
}
}Cursor shows Needs authentication next to the server. Click the button beside it to sign in.
VS Code
Install in VS Code (opens in a new tab)
Or from a terminal:
code --add-mcp '{"name":"nvidia-cuda-docs","type":"http","url":"https://api.copilot.nsight.ngc.nvidia.com/mcp/cuda-docs"}'Or commit it to the repo in .vscode/mcp.json:
{
"servers": {
"nvidia-cuda-docs": {
"type": "http",
"url": "https://api.copilot.nsight.ngc.nvidia.com/mcp/cuda-docs"
}
}
}VS Code asks you to sign in the first time the server starts.
Devin Desktop
- Add it to
~/.config/devin/mcp_config.json(macOS and Linux) or%APPDATA%\devin\mcp_config.json(Windows):
{
"mcpServers": {
"nvidia-cuda-docs": {
"serverUrl": "https://api.copilot.nsight.ngc.nvidia.com/mcp/cuda-docs"
}
}
}- Refresh the MCP server list in Cascade and sign in when asked.
Devin Desktop is the new name for Windsurf. It reads serverUrl (or url) for remote servers.
Codex
codex mcp add nvidia-cuda-docs --url https://api.copilot.nsight.ngc.nvidia.com/mcp/cuda-docs
codex mcp login nvidia-cuda-docsOr edit ~/.codex/config.toml directly:
[mcp_servers.nvidia-cuda-docs]
url = "https://api.copilot.nsight.ngc.nvidia.com/mcp/cuda-docs"Gemini CLI
gemini mcp add --transport http nvidia-cuda-docs https://api.copilot.nsight.ngc.nvidia.com/mcp/cuda-docsThen, inside Gemini CLI, run /mcp auth nvidia-cuda-docs to sign in.
This adds it to the current project. Add -s user to use it everywhere.
Any client
Most clients accept this shape. Some name the URL field differently: serverUrl in Devin Desktop, httpUrl in Gemini CLI’s settings file.
{
"mcpServers": {
"nvidia-cuda-docs": {
"type": "http",
"url": "https://api.copilot.nsight.ngc.nvidia.com/mcp/cuda-docs"
}
}
}Zed puts servers under context_servers in its settings. Cline needs "type": "streamableHttp", or it assumes SSE.
Client only starts local servers? Bridge it with npx -y mcp-remote https://api.copilot.nsight.ngc.nvidia.com/mcp/cuda-docs.
Embed
Show NVIDIA CUDA Docs in a README or on your site. Each embed links to this page, uses the server’s own URL, and sets no cookies. How embeds work
The card, the iframe and the JSON are in English. Add ?lang=it, fr, de or es to their address for another language.
README badge
[](https://mcp.tc/i/nvidia-cuda-docs)<a href="https://mcp.tc/i/nvidia-cuda-docs"><img src="https://mcp.tc/i/nvidia-cuda-docs/badge.svg" alt="NVIDIA CUDA Docs on mcp.tc" height="20"></a>It follows the reader’s light or dark mode. Add ?theme=light or ?theme=dark to fix it, or ?style=compact to show the mark without the word.
Website card
<script src="https://mcp.tc/w/nvidia-cuda-docs.js" async></script>The card appears where the tag is. Add data-theme="dark" or data-size="compact" to the tag to change it. The file holds everything it needs, so it makes no other requests and sets no cookies.
iframe
<iframe src="https://mcp.tc/embed/nvidia-cuda-docs" title="NVIDIA CUDA Docs on mcp.tc" width="420" height="200" loading="lazy" allow="clipboard-write" style="border:0;border-radius:8px;max-width:100%"></iframe>Use it on pages that don’t allow scripts. Add ?theme=light or ?theme=dark to the address to fix the colors.
JSON
The listing as data for your own pages and tools: name, tagline, the server’s own URL, tools and the “Add to” links. Any site can read it (CORS is open).
https://mcp.tc/i/nvidia-cuda-docs.jsonAI prompt
Paste this into an AI coding assistant, such as Claude Code or Cursor, and it adds the NVIDIA CUDA Docs card to your site.
Add the mcp.tc card for NVIDIA CUDA Docs, an MCP server, to my website. It is one static script: no build step, no package to install, no cookies.
Put this tag where the card should appear. The card renders right before it:
```html
<script src="https://mcp.tc/w/nvidia-cuda-docs.js" async></script>
```
To show the card somewhere else, put this placeholder there and load the same script once, for example before `</body>`. Keep the link inside: visitors without JavaScript see it.
```html
<div data-mcptc="nvidia-cuda-docs"><a href="https://mcp.tc/i/nvidia-cuda-docs">NVIDIA CUDA Docs on mcp.tc</a></div>
```
Options:
- `data-theme="light"` or `"dark"` on the tag or the placeholder fixes the colors. Without it, the card follows the visitor’s light or dark mode.
- `data-size="compact"` gives a smaller card without the tagline.
- The CSS variable `--mcptc-max-width` on the placeholder or a parent sets the width (default `520px`).
- `?lang=` on the script URL sets the card’s language: `en`, `it`, `fr`, `de` or `es`.
If the site has a Content Security Policy, allow `script-src https://mcp.tc` and `img-src data:` (the icon is inline).
For a README or other Markdown docs, use the badge instead:
```markdown
[](https://mcp.tc/i/nvidia-cuda-docs)
```
Reference: https://mcp.tc/embedChange history
Its tools are behind a sign-in, so we can’t read them. Edits by its owner show up here.
Nothing recorded yet. Edits by the server’s owner will show up here.
FAQ
Can I paste mcp.tc/i/nvidia-cuda-docs into my MCP client?
No. mcp.tc links are pages, not server addresses. Connect with https://api.copilot.nsight.ngc.nvidia.com/mcp/cuda-docs, so your client talks to NVIDIA CUDA Docs directly. The sign-in token only works there anyway. The quick link is for sharing: it opens this page, with setup steps for every client.
Does NVIDIA CUDA Docs need an API key or a sign-in?
Yes, you need a NVIDIA account. The first time your client connects, it opens NVIDIA’s sign-in page. The token goes straight to the server’s own URL, never through mcp.tc.
Is NVIDIA CUDA Docs a remote or a local server?
Remote. NVIDIA hosts it at https://api.copilot.nsight.ngc.nvidia.com/mcp/cuda-docs, and it speaks Streamable HTTP. There’s nothing to install.
What can NVIDIA CUDA Docs do?
You can search NVIDIA CUDA documentation, find CUDA code samples and examples and ground CUDA answers in current first-party sources.
Which clients can use it?
Any client that supports remote MCP servers: Claude Code, Claude Desktop, claude.ai, ChatGPT, Cursor, VS Code, Devin Desktop, Codex, Gemini CLI, Zed and others. The setup steps cover each one.
Does mcp.tc track changes to NVIDIA CUDA Docs?
Partly. NVIDIA CUDA Docs’s tools are behind a sign-in, so its History tab shows only edits by its owner. Follow the listing to get an email when it changes.
Who wrote this page?
mcp.tc’s robot read NVIDIA CUDA Docs’s own metadata (its MCP handshake, tool list and public pages), and an AI model drafted the text from it. A person reviews anything the checks can’t confirm. The text can still be wrong, so if you spot a mistake, use Report this listing on this page.
mcp.tc isn’t affiliated with NVIDIA. This page was built from NVIDIA CUDA Docs’s public metadata, last checked on 17 Nov 2026, and an AI model wrote the description, so it can be wrong. Names and marks belong to their owners. Something off? Report this listing.