Brosh

Provides an MCP-accessible interface to capture web content as screenshots, text, and HTML for AI workflows.

python

3

GitHub Stars

python

Language

5 months ago

First Indexed

2 months ago

Catalog Refreshed

Documentation & install

Readme and setup notes from the catalogue, plus a client-ready config you can copy for your MCP host.

Installation

Add the following to your MCP client configuration file.

Configuration

View docs

{
  "mcpServers": {
    "twardoch-brosh": {
      "command": "uvx",
      "args": [
        "brosh-mcp"
      ],
      "env": {
        "FASTMCP_LOG_LEVEL": "INFO"
      }
    }
  }
}

You can run Brosh in MCP server mode to expose a dedicated tool that lets AI agents request web captures. This enables automated retrieval of screenshots, along with optional text and HTML, all tailored for use in AI workflows and model contexts.

How to use

To use the MCP server, start one of the MCP entry points and connect your MCP client to it. You can run a local MCP server that executes Brosh captures on demand, returning images and associated metadata suitable for AI tools.

Start options include starting the dedicated MCP server script or using the main command with the MCP mode flag. You can choose between launching a small, local MCP process or running a local host command that binds to standard input/output streams for MCP communication.

On the client side, send a request containing the target URL and capture preferences (whether to include text, HTML, image data, or file paths). The server returns structured data that links each captured frame to its content region, enabling precise AI understanding and downstream processing.

How to install

Prerequisites: you need Python and a way to install Python packages, plus a mechanism to run an MCP client. You may also install a local MCP runner via a package manager that supports executable Python scripts.

Install Brosh in your environment to enable MCP server mode. You can install via a Python package manager or via a fast Python tool runner. After installation, you’ll be able to start the MCP server with the dedicated script or the main command in MCP mode.

# Basic installation using pip
python -m pip install brosh

# Optional: install all optional dependencies for MCP support
python -m pip install "brosh[all]"

# Install a fast tool runner if you want to use uvx
# (instructions vary by platform; see your environment for exact steps)

# After installation, you can run the MCP server via the dedicated entry point
brosh-mcp
# or the main command in MCP mode
brosh mcp

Other important notes

The MCP server exposes a tool named see_webpage that accepts parameters similar to capture requests, including url, width, height, scroll_step, scale, and flags to fetch HTML, text, or images. The server formats results for AI consumption, potentially returning base64-encoded images and text/HTML.

Available tools

see_webpage

MCP tool that processes capture requests and returns screenshots plus optional text and HTML suitable for AI consumption.