A flat $29/mo for 50k scrapes with no concurrency cap — or self-host the whole thing for free under Apache 2.0. No AGPL, no closed-source fire-engine. Pay-per-call coming soon.
Firecrawl pricing from firecrawl.dev/pricing (Apr 2026). Purify Pay-as-you-go is on the waitlist.
Works with your AI stack
Purify strips navigation, ads, and boilerplate before the HTML hits your model. Measured on real sites with tiktoken.
Measured using tiktoken (GPT-4 tokenizer). Raw = unmodified HTML. After Purify = extracted Markdown.
Three steps from zero to clean Markdown.
Download a single binary. No Node.js, no Python, no Docker. Just one file.
# macOS / Linuxcurl -sSL https://purify.verifly.pro/install.sh | sh # Or download directly from GitHubwget https://github.com/Easonliuliang/purify/releases/latest/download/purifyPOST any URL to the API. Works with dynamic JavaScript-heavy sites too.
curl -X POST https://purify.verifly.pro/api/v1/scrape \ -H "Authorization: Bearer YOUR_API_KEY" \ -H "Content-Type: application/json" \ -d '{"url": "https://github.com/Easonliuliang/purify"}'Receive structured, token-efficient Markdown. Ready for your LLM or AI agent.
{ "success": true, "markdown": "# Purify\n\nTurn any web page into clean Markdown...\n", "tokens_saved": "98.7%", "processing_time_ms": 420}Purify ships a built-in MCP server. Drop one config file and your AI assistant can scrape the web.
Add Purify as an MCP server in Claude Desktop or Cursor. Your AI assistant can scrape any web page on demand.
// ~/Library/Application Support/Claude/claude_desktop_config.json{ "mcpServers": { "purify": { "command": "/path/to/purify-mcp", "env": { "PURIFY_API_URL": "https://purify.verifly.pro", "PURIFY_API_KEY": "YOUR_API_KEY" } } }}No seat pricing. No concurrency wall. No closed-source anti-bot module. Everything ships in one Go binary.
Flat $29/mo for 50k scrapes with unlimited concurrency — no 5 req/s ceiling like Firecrawl's cheap tier. Pay-per-call ($0.0008) coming soon.
Apache 2.0 — the extraction engine, browser fallback, and MCP server all ship in the binary. No closed-source fire-engine. No AGPL.
One file, zero runtime dependencies. No Redis, no Postgres, no Playwright. Runs on a $5 VPS.
Native Model Context Protocol support. Connect Claude, Cursor, and other agents with one config file.
~100ms for static pages. Chromium only spins up when a page actually needs JavaScript.
Strip nav, ads, and scripts before the HTML hits your LLM. Measured on real sites with tiktoken.
The specific places the SaaS incumbents fall short — and how Purify fills each gap.
Purify is a web scraping API for AI agents. It turns any URL into clean Markdown (or a structured object via BYO LLM). Everything ships in a single Go binary you can run yourself under Apache 2.0. Today we offer a flat $29/mo plan or free self-host; pay-per-call is on the near-term roadmap.
Similar category, different tradeoffs. Firecrawl is TypeScript + 8+ services, AGPL, subscription-only, and keeps its anti-bot module (fire-engine) closed-source. Purify is a single Go binary, Apache 2.0, with the whole product fully self-hostable. Today we offer a flat $29/mo plan with no concurrency cap; pay-per-call is in development.
Agent workloads are spiky. Subscription tiers either waste money in quiet weeks or throttle you the moment traffic spikes (Firecrawl's cheapest tier caps at 5 concurrent requests). Pay-per-call removes both failure modes — and it's the next thing we're shipping. Join the waitlist on the pricing page.
Yes, completely. No Docker required, no Redis, no Postgres, no Playwright — just the binary. Self-hosted instances have no usage limits and include every feature of the managed cloud.
Model Context Protocol is an open standard that lets AI agents (Claude, Cursor, OpenClaw) call external tools. Purify's built-in MCP server lets your agent scrape any URL with one config file.
Yes. Purify tries HTTP first for speed (~100ms) and automatically falls back to a headless Chromium when the content requires JavaScript.
Join developers building smarter AI agents — start free, no credit card needed.