Nowledge

Nowledge MCP Server

This is the official Nowledge MCP Server

It takes a web URL via MCP, crawls all relevant pages, converts them to Markdown, and returns either one document or a list by page.

Features

• 🌐 Web-wide Fetching: Fetches content from any public web URL
• 🧹 HTML Sanitization: Strips headers, footers, navigation, scripts, and ads
• 🔗 Link Rewriting: Adjusts links to work in Markdown
• 📄 Multiple Output Formats: Get one document or structured pages
• 🚀 Performance: Fast crawling with adjustable concurrency and depth
• NLP: Search by library name or keyword

Usage

Prompts you can use:

nowledge fetch how can i use gpt-image-1 with "vercel ai" sdk

nowledge fetch how can i create new blocks in shadcn?

nowledge fetch i want to understand how X works

Fetch complete Documentation (Default)

use nowledge https://some-website.com/shadcn-ui/ui
use nowledge multiple pages https://some-website.com/shadcn-ui/ui

Single Page

use nowledge fetch single page https://some-website.com/tailwindlabs/tailwindcss/2.2-theme-system

Get by shortform

use nowledge fetch tailwindlabs/tailwindcss

nowledge fetch library

nowledge fetch url
nowledge fetch /

nowledge multiple pages ...
nowledge single page url ...

Cursor

Add this to .cursor/mcp.json file.

{
  "mcpServers": {
    "mcp-nowledge": {
      "command": "npx",
      "args": ["-y", "mcp-nowledge@latest"]
    }
  }
}

MCP Tool Integration

The package registers a tool named nowledge_fetch that you can use with any MCP-compatible client:

{
  "action": "nowledge_fetch",
  "params": {
    "url": "https://some-website.com/user/repo",
    "mode": "aggregate",
    "maxDepth": "1"
  }
}

Parameters

• url (required): The starting URL of the website or repository
• mode (optional): Output mode, either "aggregate" for a single Markdown document (default) or "pages" for structured page data
• maxDepth (optional): Maximum depth of pages to crawl (default: 10)

Response Format

Success Response (Aggregate Mode)

{
  "status": "ok",
  "data": "# Page Title\n\nPage content...\n\n---\n\n# Another Page\n\nMore content...",
  "totalPages": 5,
  "totalBytes": 25000,
  "elapsedMs": 1200
}

Success Response (Pages Mode)

{
  "status": "ok",
  "data": [
    {
      "path": "index",
      "markdown": "# Home Page\n\nWelcome to the repository."
    },
    {
      "path": "section/page1",
      "markdown": "# First Page\n\nThis is the first page content."
    }
  ],
  "totalPages": 2,
  "totalBytes": 12000,
  "elapsedMs": 800
}

Error Response

{
  "status": "error",
  "code": "FETCH_FAIL",
  "message": "Failed to fetch content from the web"
}

Partial Success Response

{
  "status": "partial",
  "data": "# Page Title\n\nPage content...",
  "errors": [
    {
      "url": "https://some-website.com/user/repo/page2",
      "reason": "HTTP error: 404"
    }
  ],
  "totalPages": 1,
  "totalBytes": 5000,
  "elapsedMs": 950
}

Progress Events

When using the tool, you'll receive progress events during crawling:

Fetched https://some-website.com/user/repo: 12500 bytes in 450ms (status: 200)
Fetched https://some-website.com/user/repo/page1: 8750 bytes in 320ms (status: 200)
Fetched https://some-website.com/user/repo/page2: 6200 bytes in 280ms (status: 200)

Local Development - Installation

Local Usage

{
  "mcpServers": {
    "mcp-nowledge": {
      "command": "node",
      "args": ["./bin/cli.mjs"]
    }
  }
}

From Source

# Clone the repository
git clone https://github.com/instructa/nowledge.git
cd nowledge

# Install dependencies
npm install

# Build the package
npm run build

Direct API Calls

For HTTP transport, you can make direct API calls:

curl -X POST http://localhost:3000/mcp \
  -H "Content-Type: application/json" \
  -d '{
    "id": "req-1",
    "action": "nowledge_fetch",
    "params": {
      "url": "https://some-website.com/user/repo",
      "mode": "aggregate"
    }
  }'

Configuration

Environment Variables

• NOWLEDGE_MAX_CONCURRENCY: Maximum concurrent requests (default: 5)
• NOWLEDGE_REQUEST_TIMEOUT: Request timeout in milliseconds (default: 30000)
• NOWLEDGE_MAX_RETRIES: Maximum retry attempts for failed requests (default: 3)
• NOWLEDGE_RETRY_DELAY: Base delay for retry backoff in milliseconds (default: 250)

To configure these, create a .env file in the project root:

NOWLEDGE_MAX_CONCURRENCY=10
NOWLEDGE_REQUEST_TIMEOUT=60000
NOWLEDGE_MAX_RETRIES=5
NOWLEDGE_RETRY_DELAY=500

Docker Deployment (Untested)

Build and run the Docker image:

# Build the image
docker build -t mcp-nowledge .

# Run with stdio transport (for development)
docker run -it --rm mcp-nowledge

# Run with HTTP transport (for production)
docker run -d -p 3000:3000 mcp-nowledge --http --port 3000

# Run with environment variables
docker run -d -p 3000:3000 \
  -e NOWLEDGE_MAX_CONCURRENCY=10 \
  -e NOWLEDGE_REQUEST_TIMEOUT=60000 \
  mcp-nowledge --http --port 3000

Development

# Install dependencies
pnpm install

# Run in development mode with stdio
pnpm run dev-stdio

# Run tests
pnpm test

# Run linter
pnpm run lint

# Build the package
pnpm run build

Troubleshooting

Common Issues

1. Permission Denied: If you get EACCES errors when running the CLI, make sure to make the binary executable:

``bash chmod +x ./node_modules/.bin/mcp-nowledge ``

1. Connection Refused: Make sure the port is available and not blocked by a firewall:

``bash # Check if port is in use lsof -i :3000 ``

1. Timeout Errors: For large repositories, consider increasing the timeout and concurrency:

`` NOWLEDGE_REQUEST_TIMEOUT=60000 NOWLEDGE_MAX_CONCURRENCY=10 npx mcp-nowledge ``

Contributing

We welcome contributions! Please see [CONTRIBUTING.md](CONTRIBUTING.md) for details.

License

MIT

Links

• GitHub: instructa/nowledge
• X/Twitter: @kregenrek
• Bluesky: @kevinkern.dev

Courses

• Learn Cursor AI: Ultimate Cursor Course
• Learn to build software with AI: instructa.ai

See my other projects:

• AI Prompts - Curated AI Prompts for Cursor AI, Cline, Windsurf and Github Copilot
• codefetch - Turn code into Markdown for LLMs with one simple terminal command
• aidex A CLI tool that provides detailed information about AI language models, helping developers choose the right model for their needs.

Source & license

This open-source MCP server is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: instructa
• Source: instructa/nowledge
• License: MIT

Install and usage instructions live in the source repository linked above.