Bridge MCP Server

How to use

You will run the MCP server locally or on FastMCP Cloud and connect your AI client to it. The server exposes a set of capabilities that let the AI launch and manage applications, simulate input, capture the screen, automate browser actions, run system commands, and manage the clipboard. You interact with the server through an agent that registers with your AI client and persists across sessions.

How to install

{ 
  "type": "stdio",
  "name": "bridge_mcp_local",
  "command": "python",
  "args": ["C:\\Users\\YourName\\Path\\To\\Bridge-MCP\\bridge_mcp.py"]
}

Prerequisites: you need Python installed on your system. You may also want to install additional runtime dependencies referenced by the project, such as a package manager and a web browser automation library.

Step 1: Clone the project - git clone https://github.com/BarhamAgha1/Bridge-MCP.git - cd Bridge-MCP

Step 2: Install dependencies
- pip install -r requirements-local.txt
- Note: Playwright browsers may auto-install on first use.

Step 3: Start the local MCP server - python bridge_mcp.py

Step 4: Start the local agent (Windows) - python local_agent.py

Step 5: Register your agent in your AI client using the provided agent callback URL (for example http://127.0.0.1:8006) and agent name.

Additional setup options

If you want to access your MCP server remotely, you can deploy it to FastMCP Cloud. You will receive a URL like the following where your MCP will be reachable: https://your-project.fastmcp.app/mcp. Use this URL in your client configuration to route requests to the cloud-hosted MCP.

Configuration through clients like Claude Desktop, Cursor, or VS Code can point to the local or cloud MCP endpoint. The standard approach is to configure the client to call the bridge_mcp server using the registered agent callback URL and the appropriate agent identifier.

Security considerations include token-based auth and a safety mode that requires explicit approval for dangerous actions. Ensure your local agent and MCP server are protected behind a firewall and are not exposed publicly without proper authentication.

Troubleshooting and notes

If you encounter issues, verify that the local MCP server and agent are running in their respective terminals. Check that the configured paths point to the actual bridge_mcp.py and local_agent.py files. If your client reports no agents or server disconnection, confirm registration status and reachability of the callback URL.

For cloud deployments, ensure the project has been built and deployed correctly, and that the cloud URL is used as the callback in your AI client. If you need to restart services, stop and start the local server and agent in sequence.

Notes on security and observability

Bridge MCP includes a Safety Sentinel that requires you to approve sensitive commands and provides an approval overlay in the AI activity view. You can monitor pending requests from the web dashboard and toggle Safe Mode as needed.

Live observability features include a dashboard with a high-fidelity desktop stream and visual overlays that indicate what the AI detects on the screen. This helps you verify that the AI interactions target the correct UI elements.

Semantic capabilities and resources

The server exposes resources for desktop state, screenshots, and file access, and supports prompts and session memory so your AI can maintain context across interactions.

Compliance and extension points

Bridge MCP implements a complete MCP specification, enabling rich workflows from task automation to web automation templates. You can extend it with additional tools and integrate new automation scenarios as needed.

Examples of use

Example: you can have the AI open a text editor, type content, and save the file, or you can automate browser actions to perform web searches and extract results.

What you get with Bridge MCP

A scalable, secure bridge between any MCP-compliant AI and your Windows PC, enabling automated control of applications, input, screen capture, and browser automation.

app_launch

Launch an application by name and bring it into focus.

app_switch

Switch focus to an already open application by its window title or process name.

app_close

Close a running application gracefully.

app_list

List currently open applications.

click

Simulate a mouse click at specified screen coordinates.

double_click

Perform a double-click at given coordinates.

right_click

Perform a right-click at given coordinates.

type_text

Type a given string into the focused input field.

press_key

Press a single keyboard key, such as Enter or Escape.

hotkey

Trigger a keyboard shortcut combination.

scroll

Scroll a given direction multiple steps.

drag

Drag the cursor from one coordinate to another.

move_mouse

Move the cursor to specific coordinates.

screenshot

Capture a screenshot of the current desktop.

get_desktop_state

Retrieve the full desktop state, including window layouts.

get_screen_size

Return the width and height of the screen.

get_mouse_position

Return the current cursor position.

run_powershell

Execute PowerShell commands.

run_cmd

Execute CMD commands.

file_read

Read the contents of a file.

file_write

Write data to a file.

file_list

List files and directories.

chrome_open

Open a URL in Chrome.

chrome_navigate

Navigate to a URL in Chrome.

browser_navigate

Navigate the browser to a URL using Playwright.

browser_click

Click a browser element specified by a selector.

browser_type

Type text into a browser element.

browser_press

Simulate a key press within the browser.

browser_content

Retrieve visible page text.

browser_screenshot

Capture a screenshot of the browser context.

clipboard_copy

Copy text to the system clipboard.

clipboard_paste

Retrieve text from the system clipboard.