Anybridge

anybridge

**Any website, within reach of any agent.**

anybridge turns a website into tools an AI agent can actually use. It opens the page in a real browser and re-exposes it over the **Model Context Protocol**, so Claude Code, Codex, or anything else that speaks MCP can read the page, fill its forms, click its buttons and read its PDFs — including sites built long before agents existed.

Run it, pick your agent, paste a URL into it. That is the whole flow.

Why it exists

[WebMCP](https://github.com/webmachinelearning/webmcp) lets a website declare typed, callable tools for AI agents. It is real and already deployed — Shopify ships it on storefronts today — but only Gemini in Chrome can consume those tools. anybridge captures them and hands them to **every other agent**.

And when a site has no WebMCP at all, which is almost every site, anybridge generates the bridge itself from the live DOM. A 2013 SharePoint portal and a modern Shopify store come out the same way: as a list of tools.

What an agent gets

Two sources of tools, merged, with the site's own taking precedence:

**The site's native WebMCP tools**, namespaced so a page can never shadow a built-in — `search_catalog`, `update_cart`, whatever the site registered.

**Forty universal tools that work on any page.** The ones that matter most:

navigate, read_page, smart_read	open and read, with PDFs returned as page-marked text
snapshot	a compact semantic tree with stable refs (@e12) instead of raw HTML
click_ref, fill_ref, select_ref, press_key	act on those refs
extract, screenshot, wait_for	structured data, images, synchronisation
save_site, save_profile, save_workflow	remember a site, a login, a recorded flow
open_repository	clone a Git remote and hand back a local path

Run it

One command, nothi