Feature Request: Add direct page content extraction method

[IAliceBobI]

Feature Comparison with Other MCPs

After comparing Playwright MCP with BrowserOS MCP and Chrome DevTools MCP, I noticed Playwright MCP lacks a direct page content extraction method.

Comparison Table

Feature	Playwright MCP	BrowserOS MCP	Chrome DevTools MCP
Direct content extraction	❌ Requires code	✅ get_page_content	❌ Requires JS
CSS selector filtering	❌ Requires code	✅ selector param	❌
Viewport-only extraction	❌ Requires code	✅ viewportOnly param	❌
browser_run_code	✅ Excellent!	❌	❌
browser_evaluate	✅	✅ evaluate_script	✅ evaluate_script
browser_wait_for	✅	❌ Manual polling	✅ wait_for

Missing Features in Playwright MCP

Direct page content extraction without writing code

Currently, users must write JavaScript even for simple content extraction. BrowserOS MCP provides a much simpler API with get_page_content.

Proposed API

browser_get_content: {
  selector?: string;      // CSS selector to filter content
  viewportOnly?: boolean; // Only visible content
  includeLinks?: boolean; // Format links as markdown
  fullPage?: boolean;     // Include content beyond viewport
}

Playwright MCP Strengths 🌟

• browser_run_code is incredibly powerful - allows full access to Playwright's API
• Great browser_wait_for functionality
• Excellent snapshot accessibility tree
• Proper tab management

Suggestion

Consider adding a browser_get_content method for simple content extraction use cases where writing code feels like overkill. This would complement the existing powerful browser_run_code for advanced scenarios!

[furkankoykiran]

I've implemented the browser_get_content tool as requested!

Summary:

• Added browser_get_content tool for direct page content extraction
• Supports CSS selector filtering with selector parameter
• Supports viewport-only extraction with viewportOnly parameter
• Supports multiple output formats: text, markdown, html
• Supports link formatting in markdown with includeLinks parameter

PR: microsoft/playwright#39619

This provides a convenient API for simple content extraction use cases where writing JavaScript code feels like overkill, complementing the existing powerful browser_run_code tool for advanced scenarios.

[yury-s]

@IAliceBobI do you have a practical use where you need this and existing snapshot won't do?

[IAliceBobI]

@yury-s Thanks for the follow-up! Here are my real-world use cases where snapshot falls short:

1. AI Reading Web Content (Primary Use Case)

I use Claude Code to read documentation and articles. The current browser_snapshot returns accessibility tree markers ([link], [button]) which are noisy for AI content consumption.

2. Direct Comparison

I tested both approaches against https://rust-lang.org:

Playwright MCP - browser_snapshot

Call:

mcporter call playwright.browser_navigate url: "https://rust-lang.org"
mcporter call playwright.browser_snapshot

Returns:

- generic [active] [ref=e1]:
  - navigation [ref=e2]:
    - link "Rust Logo" [ref=e4] [cursor=pointer]:
      - /url: /
      - img "Rust Logo" [ref=e5]
    - list [ref=e6]:
      - listitem [ref=e7]:
        - link "Install" [ref=e8] [cursor=pointer]:

To get actual content, I have to write JavaScript:

mcporter call playwright.browser_evaluate function: "document.body.innerText"

Returns plain text only:

Rust
A language empowering everyone
to build reliable and efficient software.
GET STARTED
...

Links and structure are lost.

---

BrowserOS MCP - get_page_content

Call:

mcporter call browserOS.new_page url: "https://rust-lang.org"
mcporter call browserOS.get_page_content page: 1 selector: "main" includeLinks: true

Returns clean markdown:

# Rust

## A language empowering everyone
 to build reliable and efficient software.

[Get Started](https://rust-lang.org/learn/get-started)
[Version 1.94.0](https://blog.rust-lang.org/releases/latest)

## Why Rust?

### Performance
Rust is blazingly fast and memory-efficient: with no runtime or
garbage collector, it can power performance-critical services...

### Reliability
Rust's rich type system and ownership model guarantee memory-safety...

Links preserved as [text](url) - AI can reference sources directly.

3. Why This Matters

Feature	browser_snapshot	browser_get_content
Format	Accessibility tree	Clean markdown
Links	[ref=e4] reference	[text](url) clickable
Readability	Good for testing	Good for AI consumption
Code required	Write JS for content	One-line call

4. Real Use Cases

Content extraction for LLM context:

# Before: Need to write JavaScript
mcporter call playwright.browser_run_code code: "async (page) => {
  const el = await page.$('article');
  return el ? await el.evaluate(e => e.innerText) : '';
}"

# After: Simple one-liner
mcporter call browser_get_content selector: "article" includeLinks: true

Research/Documentation gathering:
When I ask AI to summarize a technical article, I want:

• Clean text without UI noise (nav bars, buttons, ads)
• Preserved links for source citation
• No JavaScript coding required

@furkankoykiran Thanks for implementing browser_get_content! This complements browser_run_code perfectly - simple tasks stay simple, complex tasks stay possible.