Skip to main content
Glama
Sign In
mcp-textbrowser

mcp-textbrowser

by bytesbrains
OverviewSchemaSourceRelated ServersDiscussionsAdmin
TypeScript
Local

@bytesbrains/mcp-textbrowser

npm version license CI

MCP server — text-first headless browser for Claude Code and any MCP host. DOM + OCR text maps. Zero image tokens by default. 5-15x cheaper than screenshot-based browser MCPs.

browser_navigate(url)             → DOM elements + OCR text   (~200 tokens)
browser_navigate(url, visual=true) → text + PNG               (use for layout/color only)

Why

Every screenshot-based browser MCP sends a PNG to the AI on every action. At 1280×800 that's ~1,300 image tokens per page — 5-15x more expensive than reading the same content as text.

mcp-textbrowser captures a screenshot for OCR, extracts the text, then discards the image. Only structured DOM elements and OCR text reach the model. Switch to visual=true only when you genuinely need pixels (layout checks, color, design review).

Mode

Tokens per action

When to use

text-only (default)

~150–400

Everything: navigation, forms, data extraction, workflows

visual=true

~1,500–3,000

Layout, colors, CSS, design review

Related MCP server: MCP Master Puppeteer

Install

Step 1 — install Chromium (one-time, ~130MB):

npx playwright install chromium

Step 2 — add the MCP server:

Claude Code (CLI) — one command

claude mcp add textbrowser -- npx -y @bytesbrains/mcp-textbrowser

That's it. Restart Claude Code and the tools are ready.

Claude Desktop

Add to ~/Library/Application Support/Claude/claude_desktop_config.json (Mac) or %APPDATA%\Claude\claude_desktop_config.json (Windows):

{
  "mcpServers": {
    "textbrowser": {
      "command": "npx",
      "args": ["-y", "@bytesbrains/mcp-textbrowser"]
    }
  }
}

Restart Claude Desktop.

Manual (any MCP host)

Use command: npx, args: ["-y", "@bytesbrains/mcp-textbrowser"] in your host's MCP server config.

Tools

Tool

What it does

browser_navigate

Open a URL, return page context

browser_click

Click by CSS selector / XPath / visible text

browser_type

Fill an input field

browser_scroll

Scroll page or element into view

browser_screenshot

Capture current page context

browser_read

Read current page without navigating

browser_evaluate

Run JS in the page (safe DOM ops only)

All tools default to text-only — pass visual: true to any tool to also receive the PNG.

Example output

Page: https://example.com/
Title: Example Domain
Viewport: 1280x800

Elements (3 interactive of 14 total):
  [1] <a> href="https://iana.org/domains/example" bbox=(133,175,254,195)
      text: "More information..."

OCR (full page screenshot):
Example Domain
This domain is for use in illustrative examples in documents.
You may use this domain in literature without prior coordination or asking for permission.

Requirements

  • Node.js 18+

  • Chromium: npx playwright install chromium (one-time, ~130MB)

License

MIT © BytesBrains

Built by Agent, for Agents 🤖

Built and maintained by BytesBrains — AI automation & agents, engineered to production standards. The model proposes, code guarantees.

Available Tools

7 tools
browser_clickA

Click an element on the current page. Identify the target by CSS selector, XPath, or visible text. Returns updated page context after the click.

Inputs: text, xpath, visual, selector.

browser_evaluateA

Execute JavaScript in the current page context and return the result. Restricted to safe DOM read operations — eval, Function, fetch, WebSocket, and document.write are blocked.

Inputs: script (required).

browser_navigateA

Open a URL in the browser and return the page context (DOM elements + OCR text). Default text-only mode — screenshot is captured only for OCR then discarded, zero image tokens. Set visual=true to also receive the PNG.

Inputs: url (required), visual, headless.

browser_readA

Read the current page context (DOM elements + OCR text) without navigating or clicking anything.

Inputs: visual.

browser_screenshotA

Capture the current page and return an OCR-based text map. The screenshot itself is discarded unless visual=true.

Inputs: visual.

browser_scrollA

Scroll the current page. Use direction (up/down/left/right) with an optional pixel amount, or provide a selector to scroll that element into view.

Inputs: amount, visual, selector, direction.

browser_typeA

Type text into an input field on the current page. By default clears the field first. Returns updated page context.

Inputs: text (required), selector (required), clear, visual.

Tool Schema Changelog

Recent tool additions, removals, and schema changes observed during successful MCP inspections.

Loading tool schema changelog…

    Deploy Server
    One endpoint for every MCP toolRayrun runs every MCP server behind one URL and holds the credentials. Each call is decided per client, tool, and argument, then recorded.ray.runAd

    TDQS

    A4.1/5.0

    Scored across 7 tools

    Disambiguation5/5

    Each tool targets a distinct browser action (navigate, click, type, scroll, read, screenshot, evaluate) with no overlap in purpose.

    Naming Consistency5/5

    All tool names follow the consistent 'browser_verb' pattern, making the set predictable and easy to understand.

    Tool Count5/5

    7 tools is well-scoped for a browser automation server, covering essential interactions without being excessive or insufficient.

    Completeness4/5

    Covers core browsing actions well, but lacks explicit back/forward navigation or page refresh, which are minor gaps for typical workflows.

    Resources

    Related MCP Connectors

    Related MCP Servers

    • Fetch Browser

      TheSethRose
      A
      license
      B
      quality
      F
      maintenance
      A headless browser MCP server that allows AI agents to fetch web content and perform Google searches without API keys, supporting various output formats like Markdown, JSON, HTML, and text.
      2
      24
      MIT

    • MCP Master Puppeteer

      flrngel
      A
      license
      Not graded
      quality
      D
      maintenance
      An advanced MCP server for browser automation using Puppeteer, specifically optimized for token efficiency through minimal data returns and progressive enhancement. It enables agents to navigate pages, capture LLM-optimized screenshots, extract structured content, and perform batch interactions.
      3
      MIT
    • Plasmate

      Plasmateofficial

      plasmate-labs
      A
      license
      Not graded
      quality
      D
      maintenance
      Agent-native headless browser for AI agents. Converts web pages to a Semantic Object Model (SOM) instead of raw HTML — 17x average token reduction across real-world sites (up to 117x on complex pages). Native MCP server with fetch_page, extract_text, extract_links, and full browser automation. No API key required.
      51 npm
      Apache 2.0