Nodus Compute

Nodus + your coding agent

Connect Nodus to your agent.

Give your agent the tools to run GPU workloads, follow progress, and get results. Keep working where you already build.

Let your agent set it up

Choose your agent. Sign in. Connected.

Connect directly to Nodus, then approve access in your browser. New account? Verify your email and card to create an organization. Card setup does not collect a payment.

Claude Code

  1. Add Nodus

    Paste this into your terminal.

    Terminal
    claude mcp add --scope user --transport http nodus https://api-next.nodus-compute.ai/mcp

  2. Sign in

    In Claude Code, run /mcp, choose nodus and authenticate. Approve access in your browser.

  3. Try it

    Ask “List my Nodus workloads.” This checks your connection without starting paid compute.

Codex

  1. Add Nodus in the app

    Open Settings → MCP servers → Add server. Use nodus as the name and choose Streamable HTTP. Paste this server URL.

    Server URL
    https://api-next.nodus-compute.ai/mcp

  2. Sign in

    Save the server, select Restart, then Authenticate for nodus. Approve access in your browser.

  3. Try it

    Start a new conversation and ask “List my Nodus workloads.” This checks your connection without starting paid compute.

Using the Codex CLI?

Paste these commands into your terminal, finish browser sign-in, then start a new Codex session.

Terminal
codex mcp add nodus --url https://api-next.nodus-compute.ai/mcp
codex mcp login nodus
Cursor

  1. Add Nodus

    Open Cursor with the button below, then confirm the Nodus server in Cursor.

    Add to Cursor

  2. Sign in

    In Cursor’s MCP settings, find nodus and select Connect. Approve access in your browser, then return to Cursor.

  3. Try it

    Ask “List my Nodus workloads.” This checks your connection without starting paid compute.

Cursor didn’t open?

Open Cursor’s MCP settings and add a remote server named nodus with this URL.

Server URL
https://api-next.nodus-compute.ai/mcp

If your version opens mcp.json, merge this entry with your existing servers.

MCP configuration
{
  "mcpServers": {
    "nodus": {
      "type": "http",
      "url": "https://api-next.nodus-compute.ai/mcp"
    }
  }
}

Already added Nodus? Reconnect it in Cursor’s MCP settings. You do not need to add a second server.

Other agents and cloud workflows

Add a remote HTTP MCP server using this URL, then follow your client's sign-in prompt.

MCP URL
https://api-next.nodus-compute.ai/mcp

Your client must support HTTP MCP and OAuth. Download the configuration.

Manage connected agents. Keep your existing connection if it already works.

Local installation

One command. Your agents connected.

Run it once, choose your agents, and sign in in your browser. Setup installs the tools and skills, preserves your other settings, and checks the connection.

Install Nodus
Terminal
curl -fsSL https://nodus-compute.ai/install | sh
PowerShell
irm https://nodus-compute.ai/install.ps1 | iex

Then restart your agent and ask “List my Nodus workloads.” How setup works

Manual setup and other options

Sign in once

Run this in your terminal and finish sign-in in your browser. Every client on this machine can reuse your Nodus login.

Terminal
uvx --from 'nodus-compute[mcp]==0.11.0' nodus login

Requires uv. It installs the Nodus package for you. Already signed in? Continue below.

Choose your agent

Coding agent

Run this in your terminal. Nodus will be available across your projects.

Terminal
claude mcp add --scope user --transport stdio nodus -- uvx --from 'nodus-compute[mcp]==0.11.0' nodus-mcp

Run this in your terminal, then start a new Codex session.

Terminal
codex mcp add nodus -- uvx --from 'nodus-compute[mcp]==0.11.0' nodus-mcp

Open Cursor to review and add the Nodus tools.

Add local server to Cursor
Or copy the configuration

Merge this into ~/.cursor/mcp.json. Keep your other servers.

~/.cursor/mcp.json
{
  "mcpServers": {
    "nodus": {
      "command": "uvx",
      "args": [
        "--from",
        "nodus-compute[mcp]==0.11.0",
        "nodus-mcp"
      ]
    }
  }
}

Merge this into .vscode/mcp.json, then enable Nodus in your chat tools.

.vscode/mcp.json
{
  "servers": {
    "nodus": {
      "type": "stdio",
      "command": "uvx",
      "args": [
        "--from",
        "nodus-compute[mcp]==0.11.0",
        "nodus-mcp"
      ]
    }
  }
}

Merge this into your Gemini settings, then restart Gemini CLI.

~/.gemini/settings.json
{
  "mcpServers": {
    "nodus": {
      "command": "uvx",
      "args": [
        "--from",
        "nodus-compute[mcp]==0.11.0",
        "nodus-mcp"
      ]
    }
  }
}

Merge this into opencode.json in your project, then restart OpenCode.

opencode.json
{
  "mcp": {
    "nodus": {
      "type": "local",
      "command": [
        "uvx",
        "--from",
        "nodus-compute[mcp]==0.11.0",
        "nodus-mcp"
      ],
      "enabled": true
    }
  }
}

Add a local stdio server in your agent’s MCP settings. Use this configuration when your client accepts mcpServers.

MCP configuration
{
  "mcpServers": {
    "nodus": {
      "command": "uvx",
      "args": [
        "--from",
        "nodus-compute[mcp]==0.11.0",
        "nodus-mcp"
      ]
    }
  }
}

Works with clients that run local MCP servers, including Claude Desktop, Windsurf and Cline. Find your client’s settings or download the configuration.

Already using the Nodus plugin? You already have these tools. Plugins also bundle setup and workload skills.

Ask your agent

Reload your agent’s MCP connection or start a new session, then send:

Prompt
List my Nodus workloads.

This checks the connection without starting paid compute. For a run, give your agent your command and environment.

Prefer to ask your agent?

Copy one prompt.
Let it handle setup.

Your agent can read the instructions and help configure its connection. You complete browser sign-in.

Setup prompt
Read https://nodus-compute.ai/connect.md and help me connect Nodus to this agent. Reuse any existing Nodus connection. Verify setup by listing my workloads. Do not start paid compute.

Your first result

Connected. Now run something.

01 / Check the connection

Ask your agent to identify your organization and list Jobs. An empty list is normal for a new project. This check starts no compute.

Verify setup

02 / Review your first run

Start with a small training script, a $1 limit, and a 10-minute timeout. Review the estimate and approve the request before it submits paid compute.

Open the complete example

03 / Retrieve the result

Follow the run to its final status, then download metrics.json. A submitted Job is only the start; check the output before reporting success.

Get your output

Example limits are not a price or completion guarantee. Hosted MCP cannot read files from your laptop; use the CLI to upload source or run an image that includes your code.

Built for the tools you use

One connection. More ways to build.

Skills for your agent

Add Nodus setup and workload guidance to agents that support skills. Select your agent when prompted.

Terminal
npx skills add nodus-compute/Nodus-sdk-python

Requires Node.js. Skills add instructions. Connect the MCP tools above separately.

About skills
Need help connecting?

Run sign-in on the machine where the MCP server runs, including remote development environments. If your agent cannot find uvx, restart it after installing uv or use the full executable path. Merge configuration with existing settings. Keep API keys out of chat and configuration files. More troubleshooting.

Hosted connections use browser sign-in. If access expires, authenticate Nodus again in your client. The local installer remains available for stdio clients and remote development environments.