# Hyouka Cloud Browser MCP

Standalone remote browser and Firecrawl-compatible MCP for Vercel.

## From scratch

- MCP: @modelcontextprotocol/sdk
- Browser: playwright-core
- Chromium: @sparticuz/chromium
- Hosting: Vercel Node.js Function
- No Skyvern
- No Browser Use
- No Browserless
- No Browserbase
- No external browser API
- No local Chrome
- No local PC

## Core MCP tools

The original browser MCP remains available at /mcp.

Firecrawl-compatible MCP is available at /firecrawl-mcp and exposes scrape, batch_scrape, search, map, crawl, parse, extract, interact, agent, and monitor operations.

## Firecrawl-compatible HTTP API

/v2/scrape
/v2/batch/scrape
/v2/search
/v2/map
/v2/crawl
/v2/parse
/v2/extract
/v2/agent
/v2/monitor

The implementation is independent of Firecrawl and Crawl4AI and does not require a paid LLM.

## Authentication

Set MCP_API_KEY in Vercel to require x-api-key or Authorization: Bearer authentication. Leave it unset for public access.

## Important capability boundaries

JavaScript-rendered pages, crawling, search, structured extraction, screenshots, PDF rendering, browser interaction, caching, monitoring, and bot-protection signal detection are implemented.

No claim is made that the service bypasses CAPTCHAs or anti-bot controls. stealth and enhanced proxy labels remain capability descriptors unless a real proxy provider is configured.

Semantic LLM extraction is intentionally not required. The local JSON schema projection is heuristic.

Vercel free-plan and function limits still apply.