Rag setup project

Skill butchokoy25/lightrag-claude-skills/skills/rag-setup-project

Persistent memory for Claude Code via LightRAG knowledge graphs — 7 skills + 2 auto-hooks

Install

npx -y skills add butchokoy25/lightrag-claude-skills --skill rag-setup-project

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

One thing to look at

2 stars2 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

What its author says it does

Copied from the file, not written here

Set up LightRAG persistent memory for a Cursor project with MCP configuration

SKILL.md

3.5 KB, as published. Nobody here has run it

RAG Setup Project — Configure LightRAG for a Cursor Project

Set up a project-level LightRAG knowledge graph instance for the current Cursor project. Creates the MCP configuration so Claude Code can query and store project-specific memory.

What this does

Creates .cursor/mcp.json in the project root with the LightRAG MCP server entry
Merges with existing MCP config if other servers are already configured
Optionally seeds the graph with project context (README, package.json, architecture docs)

Setup steps

1. Create or merge MCP config

If .cursor/mcp.json does NOT exist:

node -e "
const fs = require('fs');
const config = {
  mcpServers: {
    'lightrag-projects': {
      command: 'npx',
      args: ['-y', 'lightrag-mcp@latest'],
      env: {
        LIGHTRAG_URL: 'http://YOUR_LIGHTRAG_HOST:YOUR_PROJECT_PORT',
        LIGHTRAG_API_KEY: process.env.LIGHTRAG_API_KEY || 'YOUR_API_KEY'
      }
    }
  }
};
fs.mkdirSync('.cursor', { recursive: true });
fs.writeFileSync('.cursor/mcp.json', JSON.stringify(config, null, 2));
console.log('Created .cursor/mcp.json');
"

If .cursor/mcp.json already exists, merge:

node -e "
const fs = require('fs');
const existing = JSON.parse(fs.readFileSync('.cursor/mcp.json', 'utf-8'));
existing.mcpServers = existing.mcpServers || {};
existing.mcpServers['lightrag-projects'] = {
  command: 'npx',
  args: ['-y', 'lightrag-mcp@latest'],
  env: {
    LIGHTRAG_URL: 'http://YOUR_LIGHTRAG_HOST:YOUR_PROJECT_PORT',
    LIGHTRAG_API_KEY: process.env.LIGHTRAG_API_KEY || 'YOUR_API_KEY'
  }
};
fs.writeFileSync('.cursor/mcp.json', JSON.stringify(existing, null, 2));
console.log('Merged lightrag-projects into existing .cursor/mcp.json');
"

2. Update .gitignore

Suggest adding .cursor/ to .gitignore if not already present:

node -e "
const fs = require('fs');
const gitignore = fs.existsSync('.gitignore') ? fs.readFileSync('.gitignore', 'utf-8') : '';
if (!gitignore.includes('.cursor')) {
  fs.appendFileSync('.gitignore', '\n# Cursor IDE config\n.cursor/\n');
  console.log('Added .cursor/ to .gitignore');
} else {
  console.log('.cursor/ already in .gitignore');
}
"

3. Optionally seed the graph

If the user wants to seed the project graph with existing docs:

node -e "
const fs = require('fs');
const BASE = 'http://YOUR_LIGHTRAG_HOST:YOUR_PROJECT_PORT';
const API_KEY = process.env.LIGHTRAG_API_KEY || 'YOUR_API_KEY';

const files = ['README.md', 'package.json', 'ARCHITECTURE.md', 'CLAUDE.md']
  .filter(f => fs.existsSync(f));

(async () => {
  for (const file of files) {
    const text = fs.readFileSync(file, 'utf-8');
    await fetch(BASE + '/documents/text', {
      method: 'POST',
      headers: {
        'Content-Type': 'application/json',
        'X-API-Key': API_KEY
      },
      body: JSON.stringify({
        file_source: '[SEED] ' + file + ' — project bootstrap',
        text: text.slice(0, 10000)
      })
    });
    console.log('Seeded:', file);
  }
  console.log('Done. Seeded ' + files.length + ' files.');
})();
"

After setup

Restart Claude Code to pick up the new MCP config
Use /rag-project-query to query project memory
Use /rag-project-remember to store project-specific knowledge
Use /rag-project-sync at end of session to persist learnings