Skip to main content

Documentation Index

Fetch the complete documentation index at: https://docs.nudj.cx/llms.txt

Use this file to discover all available pages before exploring further.

Model Context Protocol (MCP) lets AI assistants talk securely to your Nudj organisation. This page covers the admin-panel flow: mint an MCP token, then wire up Claude Desktop (or another stdio-based client) to use it.
For Claude Code (the CLI) see Claude Code Setup. For the full tool catalog see MCP Server.

​
Step 1: Generate an MCP token

  1. Open the admin panel at https://<your-subdomain>.nudj.cx/admin.
  2. Navigate to Settings → Developer → MCP Tokens.
  3. Click Generate Token.
  4. Name the token (e.g. “My Claude Desktop”).
  5. Pick an RBAC scope — the token inherits the role’s permissions:
    • Viewer: read-only across the MCP tool surface.
    • Moderator / Manager: read + targeted writes.
    • Admin / SuperAdmin: full read-write on Nudj resources.
  6. Copy the token immediately — it’s displayed once.
MCP tokens grant your RBAC-scoped privileges to whichever AI agent you hand them to. Pick the lowest scope that still lets the agent do useful work.

​
Step 2: Configure Claude Desktop

Claude Desktop uses a stdio transport — it spawns mcp-remote as a child process. Edit the Claude Desktop config file:
  • macOS: ~/Library/Application Support/Claude/claude_desktop_config.json
  • Windows: %APPDATA%\Claude\claude_desktop_config.json
Add the nudj server: 
{
  "mcpServers": {
    "nudj": {
      "command": "npx",
      "args": [
        "-y",
        "mcp-remote",
        "https://mcp.nudj.cx/sse",
        "--header",
        "Authorization: Bearer YOUR_TOKEN_HERE",
        "--header",
        "x-api-domain: https://YOUR_SUBDOMAIN.nudj.cx/api/v2/admin"
      ]
    }
  }
}
Replace YOUR_TOKEN_HERE and YOUR_SUBDOMAIN with the values from step 1. Restart Claude Desktop. The Nudj tools should appear in the MCP panel.

​
Step 3: Start chatting

Ask Claude natural-language questions and it will invoke the MCP tools to answer:
“Which of my challenges completed the most times this week?” “Create three quiz challenges about space travel, each worth 50 points, with a ‘Science Fans’ tag.” “Summarise points distributed in the last 30 days and flag any unusual spikes.”

​
Security & safety

RBAC-scoped

The token can only do what your role allows. A Viewer token fails 403 on mutations.

Revoke instantly

Delete the token from Settings → Developer → MCP Tokens. The AI loses access within seconds.

Org-scoped

Tokens are locked to a single organisation — they won’t work against other orgs, even on accident.

No training

Your data is used to answer your prompts in that session. Anthropic does not use MCP traffic to train models.

​
Common questions

The AI immediately loses access to your Nudj community. You’ll need to generate a new token and update the Claude Desktop config to reconnect.
Yes — generate one token per assistant or teammate so you can track and revoke them independently.
No. Any client that speaks MCP works. For the Claude Code CLI, see Claude Code Setup. For a custom HTTP client, see Bearer + Domain Headers.

MCP Server Overview

The full tool catalog and transport details.

Claude Code Setup

Developer-side .mcp.json for the Claude Code CLI.