Works with Claude Code, Cursor, Codex & more

Turn any website

into an

MCP server.

Scrape pages, browse interactively, and create custom automations — all through MCP. Group your tools into servers and connect each client to exactly what it needs.

claude code
Check the price of the Sony WH-1000XM5 on Amazon

I'll use your amazon-tools server to check that.

Ran check-price(product="WH-1000XM5")

Sony WH-1000XM5

Price: $278.00 (was $399.99)

In stock · Prime eligible · 4.6★ (12,847 ratings)

The XM5 is at $278, 30% off the $399.99 list price. That's near its all-time low.

Three ways to give your agent web access

Read pages, browse interactively, or create custom automations your agent can call by name.

Scrape & read

Call map_site(url) to discover pages, then scrape_page() to read content as clean markdown. Cached and fast.

map_sitescrape_page

Interactive browsing

Open a real browser session. Click, type, scroll, extract data, and take screenshots. Captchas handled automatically.

start_sessionobserveactextract_datascreenshot

Vibe tools

Create reusable browser automations with natural language. They appear as native MCP tools your agent can call by name. Test-validated on create.

create_vibe_toolyour custom tools

One account, many servers

Group your vibe tools into named servers. Each server gets a unique MCP endpoint URL — connect different clients to exactly the tools they need.

1

Sign in

Google OAuth. No API keys. One account for everything.

2

Create servers

Name them by project or use case. "Amazon Tools", "Docs Reader", etc.

3

Build vibe tools

Call create_vibe_tool() with a natural language instruction. It's test-run before saving.

4

Connect clients

Add each server's endpoint URL to the MCP client that needs it. Done.

Why not just let your agent fetch pages itself?

Because raw fetching is slow, noisy, and breaks on half the web. site2mcp gives your agent clean data, real browser sessions, and reusable automations — without the mess.

Built-in agent fetch

Slow, noisy, and unreliable

Agents scrape raw HTML or render full pages on every call. Content is inconsistent, cluttered with nav and boilerplate, and there is no caching. Every request is a fresh slow round-trip that may return completely different results.

Site-specific MCPs

Only exist for a handful of sites

A few popular tools publish their own MCP servers. The other 99% of the web does not. You should not have to build a custom MCP for every site your agent needs to read.

CLAUDE.md / system prompt

Pollutes context and gets forgotten

Pasting documentation into CLAUDE.md floods every single request with thousands of tokens your agent did not ask for. Claude frequently ignores or forgets instructions buried deep in a bloated context window.

site2mcp

Scrape, browse, and automate

Read pages as clean markdown, browse sites interactively with a real browser, and turn repeatable tasks into one-click vibe tools. Results are cached. Group tools into servers so each client sees only what it needs.

How people use site2mcp

Any site your agent needs to read or interact with — site2mcp can handle it.

Documentation & APIs

Read any docs or API reference

Point your agent at Next.js, Stripe, Tailwind, or any documentation site. It discovers pages, reads exactly what it needs, and writes integration code with the right parameters — no copy-pasting.

nextjs.org/docsstripe.com/docstailwindcss.com/docsdocs.github.com
E-commerce

Monitor prices and products

Create vibe tools that check product prices, stock levels, or reviews. Your agent calls them by name — no browser setup needed each time.

amazon.combestbuy.comebay.com
Automation

Build custom browser workflows

Turn any repeatable web task into a vibe tool. Fill forms, extract data, check status pages. Test-validated on creation so they work reliably.

Custom automationsForm fillingData extraction

Give your agent superpowers

One account. Unlimited sites. Scrape, browse, and automate — all through MCP. Sign in with Google and start in seconds.

Get Started