Skip to main content
Glama

⚑ Why Atlas?

Traditional web scraping returns megabytes of raw, unstructured HTML bloated with navigation bars, cookie banners, tracking scripts, and presentational CSS. Feeding raw HTML into Large Language Models wastes 70% to 90% of your context window, escalates inference costs, and increases reasoning hallucinations.

Atlas compiles the web at the edge before it reaches your models:

  • 🧠 0-LLM Deterministic Pipeline: Pure heuristic AST transformation passes. Zero hallucinations, zero prompt drift, 100% reproducible output.

  • πŸ“‰ 70–90% Token Reduction: Strips page chrome while faithfully preserving headings, tables, code fences, equations, and outbound links.

  • πŸ€– Native Model Context Protocol (MCP): 1-click integration with Cursor, Claude Desktop, Windsurf, VS Code, and custom AI agent loops.

  • πŸš€ Sub-1.5s Global Edge Latency: Distributed execution across 300+ edge locations with DNS-aware SSRF protection.

  • πŸ›‘οΈ Workspace-Scoped & Idempotent: Automatic duplicate prevention via RFC 4122 idempotency keys.


Related MCP server: webclaw

πŸ“¦ What’s in This Repository

This monorepo houses the public ecosystem and developer tools for Atlas:

Component

Path

Language / Standard

Installation / Package

Model Context Protocol

mcp/

MCP Standard (2026 era)

npx -y @atlascompiler/mcp init

TypeScript / JS SDK

sdk/js/

TypeScript / Node / Deno / Bun

npm install @atlascompiler/sdk

Python SDK

sdk/python/

Python 3.10+ (httpx, typing)

pip install atlascompiler

Go SDK

sdk/go/

Go 1.22+

go get go.atlas-compiler.com/sdk

Atlas Agent Skills

skills/

Agent Skill Specification

Canonical recipes for AI assistants

Starter Examples

examples/

Node, Python, Go, cURL

Runnable recipes & templates


πŸš€ Quickstarts

1. AI Agents & Coding Assistants (MCP Server)

Connect Claude Desktop, Cursor, VS Code, or Windsurf directly to Atlas with the 1-click interactive wizard:

npx -y @atlascompiler/mcp init

The wizard prompts for your Atlas API key, validates connectivity, and automatically configures your installed editors without exposing tokens in process args.

To run the Stdio server directly:

npx -y @atlascompiler/mcp

2. TypeScript / JavaScript

Install the official type-safe client library:

npm install @atlascompiler/sdk
# or: pnpm add @atlascompiler/sdk / yarn add @atlascompiler/sdk
import { AtlasClient } from '@atlascompiler/sdk';

const client = new AtlasClient({
  apiKey: process.env.ATLAS_API_KEY!,
});

// Compile a web page into clean, structured Markdown
const response = await client.compile({
  url: 'https://docs.github.com/en/rest',
});

if (response.status === 200) {
  console.log('Title:', response.value.metadata.title);
  console.log('Markdown:\n', response.value.markdown);
}

3. Python

Install the official Python client library:

pip install atlascompiler
import os
from atlascompiler import AtlasClient

with AtlasClient(os.getenv("ATLAS_API_KEY")) as client:
    response = client.compile({"url": "https://example.com/article"})
    
    if response.status == 200:
        print("Title:", response.value["metadata"]["title"])
        print("Markdown:\n", response.value["markdown"])

Async support is built-in via AsyncAtlasClient.


4. Go

go get go.atlas-compiler.com/sdk
package main

import (
	"context"
	"fmt"
	"log"
	"os"

	"go.atlas-compiler.com/sdk"
)

func main() {
	client := atlas.New(os.Getenv("ATLAS_API_KEY"))
	res, _, err := client.Compile(context.Background(), atlas.CompileUrlRequest{
		Url: "https://example.com",
	})
	if err != nil {
		log.Fatal(err)
	}

	if res.Response200 != nil {
		fmt.Println("Markdown:\n", res.Response200.Markdown)
	}
}

5. Direct HTTP / cURL

curl -X POST https://api.atlas-compiler.com/v1/compile \
  -H "Authorization: Bearer atlas_your_api_key_here" \
  -H "Content-Type: application/json" \
  -d '{"url": "https://example.com"}'

πŸ›  Supported Operations

Atlas exposes 5 core operations via REST, SDKs, and MCP:

  • compile: Deep 12-pass AST compilation with structural content graph extraction and visual formatting.

  • scrape: Sub-1.5s real-time Markdown extraction for low-latency agent loops (POST /v1/scrape).

  • map: Rapid topological spider that crawls and ranks all reachable URLs across a domain (POST /v1/map).

  • crawl: Bounded, asynchronous multi-page website crawls with path glob filtering (POST /v1/crawls).

  • batch: High-concurrency parallel retrieval for up to 500 URLs per job with unified status webhooks (POST /v1/batches).


πŸ’¬ Community & Support

  • Discord: Join our active developer community at discord.gg/z9ktRhuudq for help, agent prompts, and feature discussions.

  • GitHub Discussions: Share recipes, ask questions, or propose ideas at Discussions.

  • Bug Reports: Open an issue using our Bug Report Template.

  • Security Inquiries: Email support@atlas-compiler.com for private disclosure.


πŸ“œ Contributing & License

Contributions are welcome! Please read our Contributing Guide and Code of Conduct.

Licensed under the MIT License.

Related MCP Connectors

Related MCP Servers

  • A
    license
    Not graded
    quality
    B
    maintenance
    The web data platform for AI agents. Fetch, search, crawl, extract, monitor, and screenshot any URL. 55+ domain extractors, 65-98% token savings. 7 MCP tools included.
    332 npm
    12
    AGPL 3.0
  • A
    license
    A
    quality
    A
    maintenance
    Web content extraction for AI agents. 10 tools: scrape, crawl, map, batch, extract, summarize, diff, brand, search, research. Uses TLS fingerprinting to bypass anti-bot without a headless browser. Outputs LLM-optimized markdown with 67% fewer tokens than raw HTML.
    10
    2,383
    AGPL 3.0
  • A
    license
    A
    quality
    A
    maintenance
    Web scraping, crawling, and structured data extraction for AI agents. 5 tools: scrape (clean markdown from any URL), crawl (entire sites), map (discover URLs), extract (structured JSON), and search. 833ms avg latency, single binary, self-hostable.
    8
    1,136
    AGPL 3.0
  • A
    license
    A
    quality
    C
    maintenance
    Web tools for AI agents. Search the web for full page content, fetch URLs as clean markdown including PDFs, extract structured data from a page with a prompt, and run multi-source deep research that returns a cited report.
    4
    2
    MIT