@voiys/fetcher-mcp

@voiys/fetcher-mcp

MCP server for fetching web content via Playwright. Connects to an existing Chrome/Chromium instance via CDP.

Installation

npx @voiys/fetcher-mcp --cdp-url=ws://localhost:9222

MCP Client Configuration

{
  "mcpServers": {
    "fetcher": {
      "command": "npx",
      "args": [
        "@voiys/fetcher-mcp",
        "--cdp-url=ws://localhost:9222"
      ]
    }
  }
}

With Steel:

{
  "mcpServers": {
    "fetcher": {
      "command": "npx",
      "args": [
        "@voiys/fetcher-mcp",
        "--cdp-url=wss://connect.steel.dev?apiKey=XXX&sessionId=YYY"
      ]
    }
  }
}

Tools

fetch_url

Fetch a single URL:

await mcpClient.callTool("fetch_url", {
  url: "https://company.com/careers"
});

fetch_urls

Fetch multiple URLs in parallel:

await mcpClient.callTool("fetch_urls", {
  urls: ["https://a.com", "https://b.com"]
});

evaluate

Execute JavaScript on an existing session (requires keepSession: true on fetch_url):

// First fetch with keepSession
const result = await mcpClient.callTool("fetch_url", {
  url: "https://example.com",
  keepSession: true
});
// Extract sessionId from response

// Then interact with the page
await mcpClient.callTool("evaluate", {
  sessionId: "...",
  script: "document.querySelector('.load-more').click()"
});

// Get updated content
await mcpClient.callTool("evaluate", {
  sessionId: "...",
  script: "null",
  returnContent: true
});

close_session

Close a persistent session (or let it auto-close after 5 min):

await mcpClient.callTool("close_session", { sessionId: "..." });
// Or close all: 
await mcpClient.callTool("close_session", {});

Options

Option	Type	Default	Description
url / urls	string / string[]	required	URL(s) to fetch
stripTags	string[]	['script','style','svg','noscript','iframe','link[rel="stylesheet"]','meta']	CSS selectors to remove before processing
returnHtml	boolean	false	Return HTML instead of Markdown
structured	boolean	false	Return JSON {title, url, content, metadata} instead of plain text
timeout	number	30000	Page load timeout in ms
waitUntil	string	'load'	Navigation wait condition: 'load', 'domcontentloaded', 'networkidle'
maxLength	number	0	Max content length (0 = unlimited)
disableMedia	boolean	true	Block images, fonts, stylesheets
evaluate	string	undefined	JavaScript to execute on page before extracting content
keepSession	boolean	false	Keep session alive, returns sessionId for use with evaluate tool

Examples

Custom tag stripping:

await mcpClient.callTool("fetch_url", {
  url: "https://company.com/jobs/123",
  stripTags: ["nav", "footer", "[class*='sidebar']", "script", "style"]
});

Execute JavaScript (click button, scroll, wait for content):

await mcpClient.callTool("fetch_url", {
  url: "https://example.com/infinite-scroll",
  evaluate: `
    // Click "Load More" button
    document.querySelector('.load-more-btn')?.click();
    // Wait for content to load
    await new Promise(r => setTimeout(r, 2000));
    // Scroll to bottom
    window.scrollTo(0, document.body.scrollHeight);
  `
});

Starting Chrome with CDP

chrome --remote-debugging-port=9222

Recommended Stealth Flags

chrome --remote-debugging-port=9222 \
  --disable-blink-features=AutomationControlled \
  --disable-features=IsolateOrigins,site-per-process \
  --no-sandbox \
  --disable-setuid-sandbox \
  --disable-dev-shm-usage \
  --disable-webgl \
  --disable-infobars \
  --disable-extensions

Flag	Purpose
--disable-blink-features=AutomationControlled	Hide automation detection
--disable-features=IsolateOrigins,site-per-process	Disable site isolation
--no-sandbox	Disable sandboxing (needed in containers)
--disable-setuid-sandbox	Disable setuid sandbox
--disable-dev-shm-usage	Avoid /dev/shm issues in Docker
--disable-webgl	Reduce fingerprinting surface
--disable-infobars	Hide "Chrome is being controlled" bar
--disable-extensions	No extensions

License

Licensed under the MIT License

Originally forked from jae-jae/fetcher-mcp