🛰️ Scrape It Now!

Web scraper made for AI and simplicity in mind. It runs as a CLI that can be parallelized and outputs high-quality markdown content.

Features

Shared:

• 🏗️ Decoupled architecture with Azure Queue Storage or local sqlite
• ⚙️ Idempotent operations that can be run in parallel
• 💾 Scraped content is stored in Azure Blob Storage or local disk

Scraper:

• 🛑 Avoid re-scraping a page if it hasn't changed
• 🚫 Block ads to lower network costs with The Block List Project
• 🔗 Explore pages in depth by detecting links and de-duplicating them
• ✍️ Extract markdown content from a page with Pandoc
• 🏷️ Extract metadata elements from the page
• 🖥️ Load dynamic JavaScript content with Playwright and Chromium
• 🕵️‍♂️ Preserve anonymity with a random user agent, random viewport size, and no client hints headers
• 📊 Show progress with a status command
• 🖼️ Store images collected on the page
• 📸 Store screenshot of the page
• 📡 Track progress of total network usage

Indexer:

• 🧠 AI Search index is created automatically
• ✂️ Chunk markdown while keeping the content coherent
• 📈 Embed chunks with OpenAI embeddings
• 🔍 Indexed content is semantically searchable with Azure AI Search

Installation

From PyPI

# Install the package
python3 -m pip install scrape-it-now
# Run the CLI
scrape-it-now --help

To configure the CLI (including authentication to the backend services), use environment variables, a .env file or command line options.

From sources

Application must be run with Python 3.13 or later. If this version is not installed, an easy way to install it is pyenv.

# Download the source code
git clone https://github.com/clemlesne/scrape-it-now.git
# Move to the directory
cd scrape-it-now
# Run install scripts
make install dev
# Run the CLI
scrape-it-now --help

How to use

Scrape a website

Run a job

Usage with Azure Blob Storage and Azure Queue Storage:

# Azure Storage configuration
export AZURE_STORAGE_ACCESS_KEY=xxx
export AZURE_STORAGE_ACCOUNT_NAME=xxx
# Run the job
scrape-it-now scrape run https://nytimes.com

Usage with Local Disk Blob and Local Disk Queue:

# Local disk configuration
export BLOB_PROVIDER=local_disk
export QUEUE_PROVIDER=local_disk
# Run the job
scrape-it-now scrape run https://nytimes.com

Example:

❯ scrape-it-now scrape run https://nytimes.com
2024-11-08T13:18:49.169320Z [info     ] Start scraping job lydmtyz
2024-11-08T13:18:49.169392Z [info     ] Installing dependencies if needed, this may take a few minutes
2024-11-08T13:18:52.542422Z [info     ] Queued 1/1 URLs
2024-11-08T13:18:58.509221Z [info     ] Start processing https://nytimes.com depth=1 process=scrape-lydmtyz-4 task=63dce50
2024-11-08T13:19:04.173198Z [info     ] Loaded 154554 ads and trackers process=scrape-lydmtyz-4
2024-11-08T13:19:16.393045Z [info     ] Queued 310/311 URLs            depth=1 process=scrape-lydmtyz-4 task=63dce50
2024-11-08T13:19:16.393323Z [info     ] Scraped                        depth=1 process=scrape-lydmtyz-4 task=63dce50
...

Most frequent options are:

For documentation on all available options, run:

scrape-it-now scrape run --help

Show job status

Usage with Azure Blob Storage:

# Azure Storage configuration
export AZURE_STORAGE_CONNECTION_STRING=xxx
# Show the job status
scrape-it-now scrape status [job_name]

Usage with Local Disk Blob:

# Local disk configuration
export BLOB_PROVIDER=local_disk
# Show the job status
scrape-it-now scrape status [job_name]

Example:

❯ scrape-it-now scrape status lydmtyz
{"created_at":"2024-11-08T13:18:52.839060Z","last_updated":"2024-11-08T13:19:16.528370Z","network_used_mb":2.6666793823242188,"processed":1,"queued":311}

Most frequent options are:

For documentation on all available options, run:

scrape-it-now scrape status --help

Index a scraped website

Run a job

Usage with Azure Blob Storage, Azure Queue Storage and Azure AI Search:

# Azure OpenAI configuration
export AZURE_OPENAI_API_KEY=xxx
export AZURE_OPENAI_EMBEDDING_DEPLOYMENT_NAME=xxx
export AZURE_OPENAI_EMBEDDING_DIMENSIONS=xxx
export AZURE_OPENAI_EMBEDDING_MODEL_NAME=xxx
export AZURE_OPENAI_ENDPOINT=xxx

# Azure Search configuration
export AZURE_SEARCH_API_KEY=xxx
export AZURE_SEARCH_ENDPOINT=xxx

# Azure Storage configuration
export AZURE_STORAGE_ACCESS_KEY=xxx
export AZURE_STORAGE_ACCOUNT_NAME=xxx

# Run the job
scrape-it-now index run [job_name]

Usage with Local Disk Blob, Local Disk Queue and Azure AI Search:

# Azure OpenAI configuration
export AZURE_OPENAI_API_KEY=xxx
export AZURE_OPENAI_EMBEDDING_DEPLOYMENT_NAME=xxx
export AZURE_OPENAI_EMBEDDING_DIMENSIONS=xxx
export AZURE_OPENAI_EMBEDDING_MODEL_NAME=xxx
export AZURE_OPENAI_ENDPOINT=xxx
# Azure Search configuration
export AZURE_SEARCH_API_KEY=xxx
export AZURE_SEARCH_ENDPOINT=xxx
# Local disk configuration
export BLOB_PROVIDER=local_disk
export QUEUE_PROVIDER=local_disk
# Run the job
scrape-it-now index run [job_name]

Example:

❯ scrape-it-now index run lydmtyz
2024-11-08T13:20:37.129411Z [info     ] Start indexing job lydmtyz
2024-11-08T13:20:38.945954Z [info     ] Start processing https://nytimes.com process=index-lydmtyz-4 task=63dce50
2024-11-08T13:20:39.162692Z [info     ] Chunked into 7 parts           process=index-lydmtyz-4 task=63dce50
2024-11-08T13:20:42.407391Z [info     ] Indexed 7 chunks               process=index-lydmtyz-4 task=63dce50
...

Most frequent options are:

For documentation on all available options, run:

scrape-it-now index run --help

Architecture

Scrape

---
title: Scrape process with Azure Storage
---
graph LR
  cli["CLI"]
  web["Website"]

  subgraph "Azure Queue Storage"
    to_chunk["To chunk"]
    to_scrape["To scrape"]
  end

  subgraph "Azure Blob Storage"
    subgraph "Container"
      job["job"]
      scraped["scraped"]
      state["state"]
    end
  end

  cli -- (1) Pull message --> to_scrape
  cli -- (2) Get cache --> scraped
  cli -- (3) Browse --> web
  cli -- (4) Update cache --> scraped
  cli -- (5) Push state --> state
  cli -- (6) Add message --> to_scrape
  cli -- (7) Add message --> to_chunk
  cli -- (8) Update state --> job

Index

---
title: Scrape process with Azure Storage and Azure AI Search
---
graph LR
  search["Azure AI Search"]
  cli["CLI"]
  embeddings["Azure OpenAI Embeddings"]

  subgraph "Azure Queue Storage"
    to_chunk["To chunk"]
  end

  subgraph "Azure Blob Storage"
    subgraph "Container"
      scraped["scraped"]
    end
  end

  cli -- (1) Pull message --> to_chunk
  cli -- (2) Get cache --> scraped
  cli -- (3) Chunk --> cli
  cli -- (4) Embed --> embeddings
  cli -- (5) Push to search --> search

Design

Blob storage is organized in folders:

[job_name]-scraping/            # Job name (either defined by the user or generated)
    scraped/                    # All the data from the pages
        [page_id]/              # Assets from a page
            screenshot.jpeg     # Screenshot (if enabled)
            [image_id].[ext]    # Image binary (if enabled)
            [image_id].json     # Image metadata (if enabled)
        [page_id].json          # Data from a page
    state/                      # Job states (cache & parallelization)
        [page_id]               # Page state
    job.json                    # Job state (aggregated stats)

Page data is considered as an API (won't break until the next major version) and is stored in JSON format:

{
  "created_at": "2024-09-11T14:06:43.566187Z",
  "redirect": "https://www.nytimes.com/interactive/2024/podcasts/serial-season-four-guantanamo.html",
  "status": 200,
  "url": "https://www.nytimes.com/interactive/2024/podcasts/serial-season-four-guantanamo.html",
  "content": "## Listen to the trailer for Serial Season 4...",
  "etag": null,
  "links": [
    "https://podcasts.apple.com/us/podcast/serial/id917918570",
    "https://music.amazon.com/podcasts/d1022069-8863-42f3-823e-857fd8a7b616/serial?ref=dm_sh_OVBHkKYvW1poSzCOsBqHFXuLc",
    ...
  ],
  "metas": {
    "description": "“Serial” returns with a history of Guantánamo told by people who lived through key moments in Guantánamo’s evolution, who know things the rest of us don’t about what it’s like to be caught inside an improvised justice system.",
    "articleid": "100000009373583",
    "twitter:site": "@nytimes",
    ...
  },
  "network_used_mb": 1.041460037231445,
  "raw": "......",
  "valid_until": "2024-09-11T14:11:37.790570Z"
}

Then, indexed data is stored in Azure AI Search:

Advanced usage

Whitelist

Whitelist option allows to restrict to a domain and ignore sub paths. It is a list of regular expressions:

domain1,regexp1,regexp2 domain2,regexp3

For examples:

To whitelist learn.microsoft.com:

learn\.microsoft\.com

To whitelist learn.microsoft.com and go.microsoft.com, but ignore all sub paths except /en-us:

learn\.microsoft\.com,^/(?!en-us).* go\.microsoft\.com

Source environment variables

To configure easily the CLI, source environment variables from a .env file. For example, for the --azure-storage-access-key option:

AZURE_STORAGE_ACCESS_KEY=xxx

For arguments that accept multiple values, use a space-separated list. For example, for the --whitelist option:

WHITELIST=learn\.microsoft\.com go\.microsoft\.com

Application cache directory

The cache directoty depends on the operating system:

• ~/.config/scrape-it-now (Unix)
• ~/Library/Application Support/scrape-it-now (macOS)
• C:\Users\\AppData\Roaming\scrape-it-now (Windows)

Broswer binary installation

Browser binaries are automatically downloaded or updated at each run. Browser is Chromium and it is not configurable (feel free to open an issue if you need another browser), it weights around 450MB. Cache is stored in the cache directory.

How Local Disk storage works

Local Disk storage is used for both blob and queue. It is not recommended for production use, as it is not easily scalable, and not fault-tolerant. It is useful for testing and development or when you cannot use Azure services.

Implementation:

• Local Disk Blob uses a directory structure to store blobs. Each blob is stored in a file with the blob name as the file name. Lease is implemented with lock files. By default, files are stored in a directory relative to the command execution directory.
• Local Disk Queue uses a SQLite database to store messages. Database is stored in the cache directory. SQL databases implement visibility timeout and deletion tokens to ensure consistency to the stateless queue services like Azure Queue Storage.

Use proxies for anonymity

Proxies are not implemented in the application. Network security cannot be achieved from the application level. Use a VPN (e.g. your, third-party) or a proxy service (e.g. residential procies, Tor) to ensure anonymity and configure the system firewall to limit the application network access to it.

Bundle with a container

As the application is packaged to PyPi, it can easily be bundled with a container. At every start, the application will download the dependencies (browser, etc.) and cache them. You can pre-download them by running the command scrape-it-now scrape install.

A good technique for performance would also to parallelize the scraping and indexing jobs by running multiple containers of each. This can be achieved with KEDA, by configuring a queue scaler.