Remote Browser MCP
Provides a persistent OpenAI tunnel and supports creating a private ChatGPT plugin, enabling remote ChatGPT access to the local browser MCP runtime and existing Edge browser session.
This server does not currently have a deployable release on Glama. Follow the repository's setup instructions to install and configure it for your MCP client, including any required credentials and local dependencies.
RemoteBrowserMCP
1. Introduction
I started RemoteBrowserMCP because I wanted to use ChatGPT on the web to operate the browser on my own computer—for example, to read pages, gather information, or fill in forms.
ChatGPT can analyze links, text, and screenshots, but opening a page in my browser, clicking a button, and continuing a task requires tools that can connect to that browser. For websites where I am already signed in, I also wanted it to reuse that browser session with my permission, reducing the need to copy content back and forth or perform each step myself.
RemoteBrowserMCP provides those browser tools through MCP and uses OpenAI Secure MCP Tunnel to connect them to ChatGPT on the web. The tools run locally, and the computer initiates the Tunnel connection, so there is no need to expose the browser's debugging port to the internet.
Once connected, you can describe a task in the conversation and let ChatGPT read pages, click, and type in Edge or Chrome after you have authorized remote debugging. The project protects existing tabs by default and prefers separate background tabs to minimize interruptions to everyday browsing. Calls within the same MCP process also reuse the browser connection instead of reconnecting for every operation.
The project includes installers for Windows, Linux, and macOS, with graphical and command-line interfaces that share the same logic. Install it, configure the Tunnel, enable browser debugging, and register the plugin in ChatGPT to connect to your browser from ChatGPT on the web. The tools can also be used with other local MCP-compatible clients.
Related MCP server: Browser MCP Gateway
2. Quick Start
Install RemoteBrowserMCP, configure the connection, and verify it in ChatGPT. You need an internet connection and a ChatGPT account or workspace that supports custom MCP Tunnel connections. Installers are available for Windows x64, Linux x64, and Intel macOS.
2.1. Install
Download the installer for your system from the latest Release.
2.1.1. Graphical Interface
Windows
Run RemoteBrowserMCP-v*-windows-amd64.exe and complete the installation wizard. Open Configure Tunnel from the Start Menu.
macOS
Open RemoteBrowserMCP-v*-darwin-amd64.dmg and run the installer application. On the completion screen, select Open RemoteBrowserMCP configuration to open the configuration UI in your browser.
Linux
Run RemoteBrowserMCP-v*-linux-amd64.run from your desktop environment. If execution is blocked, enable it in the file properties. On the completion screen, select Open RemoteBrowserMCP configuration to open the configuration UI in your browser.
2.1.2. Command Line
Windows
Open a terminal in the download directory:
for %F in (RemoteBrowserMCP-v*-windows-amd64.exe) do "%~fF" --accept-licenses --default-answer --confirm-command installmacOS
Mount the DMG and run its installer:
dmg=$(find . -maxdepth 1 -name 'RemoteBrowserMCP-v*-darwin-amd64.dmg' -print -quit)
mount=$(mktemp -d)
hdiutil attach -nobrowse -readonly -mountpoint "$mount" "$dmg"
app=$(find "$mount" -maxdepth 1 -name '*.app' -print -quit)
installer=$(find "$app/Contents/MacOS" -maxdepth 1 -type f -perm -111 -print -quit)
QT_QPA_PLATFORM=offscreen "$installer" --root "$HOME/RemoteBrowserMCP" --accept-licenses --default-answer --confirm-command install
hdiutil detach "$mount"Linux
chmod +x ./RemoteBrowserMCP-v*-linux-amd64.run
QT_QPA_PLATFORM=offscreen ./RemoteBrowserMCP-v*-linux-amd64.run --root "$HOME/RemoteBrowserMCP" --accept-licenses --default-answer --confirm-command installAfter installation, configure the connection in the next section.
2.2. Configuration
Create a Tunnel ID and a Runtime API Key.
2.2.1. Graphical Interface
Enter the Tunnel ID and Runtime API Key, click Save, then Start Service. Status updates automatically; wait for Tunnel ready in the upper-right corner. To apply saved changes while the service is running, use Restart Service. A saved key is shown as a masked placeholder; leave the field blank to reuse it.
The connection page has two independent preferences:
Start local service at login registers a real startup entry for the current installation. Windows uses a per-user HKCU Run entry pointing to the bundled native GUI worker (no password stored); Linux uses a systemd user unit; macOS uses a LaunchAgent. Existing native service installations keep their service manager and change its startup setting instead. Saving never starts or stops the current service. A usable user session and permission to manage the corresponding startup entry are required; registration errors are reported rather than treated as success.
Connect Tunnel automatically controls whether a newly started service connects the saved Tunnel. It defaults to enabled for existing installations. When disabled, the service starts in standby; Connect Tunnel and Disconnect Tunnel control the current connection without changing this preference. Restart applies saved connection changes and the automatic-connection preference.
Stop Service stops this run, including the Tunnel, while retaining both preferences. If login startup remains enabled, the next login starts it again. Disabling login startup does not stop the current run. On macOS a newly enabled LaunchAgent is loaded at the next login; use Start Service to run immediately. Linux user startup normally occurs at login, not before login; this client does not enable lingering. Windows and macOS browser access likewise runs in the user's login session.
CLI equivalents (use the Windows executable path where appropriate):
./bin/remotebrowsermcp configure --id tunnel_EXAMPLE --autostart=true --auto-connect=false
./bin/remotebrowsermcp start
./bin/remotebrowsermcp connect
./bin/remotebrowsermcp disconnectThese commands reuse an already saved key. For a first setup, also pass --api-key-stdin. Omitted preference flags preserve the saved values. status reports the actual OS startup registration; an unavailable OS backend is not reported as enabled. Failed registration restores the previous connection files and attempts to restore the previous registration.
For upgrades, keep the same installation directory and use the maintenance/update flow; retain config.toml, runtime/, browser profiles and user data. Startup preferences live in private runtime/setup.json, alongside the existing Tunnel ID; the key remains in private runtime/credentials.json. Installer payloads exclude these files and data directories. Moving the installation changes its startup identity: disable the old installation's startup entry before moving, then enable it from the new location. To uninstall, first disable login startup and stop the service (or uninstall an explicitly installed native service), then remove the application.
2.2.2. Command Line
From the installation directory:
./bin/remotebrowsermcp configure
./bin/remotebrowsermcp start
./bin/remotebrowsermcp statusOn Windows, use .\bin\remotebrowsermcp.exe with the same subcommands. Interactive configure hides the Runtime API Key. For automation, use configure --id <tunnel-id> --api-key-stdin and pass the key through standard input. Use stop to stop the service.
2.2.3. Enable browser debugging
The server connects only to the local debugging endpoint at 127.0.0.1:9222. You must enable browser debugging and approve the browser's consent prompt when it appears.
Edge: in the desktop browser you want to connect, open edge://inspect/#remote-debugging, enable Remote debugging, and leave Edge running. Do not restart your personal browser just for this setup.
If you use Chrome, or Edge does not expose that switch, start a separate browser profile with loopback debugging. For example:
Linux (Google Chrome):
google-chrome --remote-debugging-address=127.0.0.1 --remote-debugging-port=9222 --user-data-dir="$HOME/.local/share/remote-browser-mcp-profile" about:blankmacOS (Google Chrome):
open -na 'Google Chrome' --args --remote-debugging-address=127.0.0.1 --remote-debugging-port=9222 --user-data-dir="$HOME/Library/Application Support/RemoteBrowserMCP/chrome-profile" about:blankOn Windows, launch your installed browser executable with the same debugging arguments and a new --user-data-dir if the browser switch is unavailable. A separate profile has separate cookies and logins; sign in only where you wish to authorize the assistant. Never expose port 9222 to the network.
2.2.4. Browser settings and local MCP
The default is Edge at http://127.0.0.1:9222. For Chrome or a custom profile, edit [browser] in private config.toml using the fields in config.example.toml. Saving Tunnel settings preserves browser and security options.
For a local stdio MCP client, no Tunnel ID or key is required. Configure the absolute bundled bin/node (Windows: bin/node.exe) as the command, and the absolute scripts/launch.mjs as its argument.
The client and installer support headless CLI operation; browser tools still require a reachable authorized Edge/Chrome debugging session. Tunnel connectivity does not establish browser readiness.
Stop any older script-managed runtime before changing to the new client. Existing encrypted credentials are not imported automatically: enter the key again. A legacy [setup] section is rejected to prevent silent migration; remove it only after stopping that runtime. Browser and security settings can remain in place.
2.3. Verify in ChatGPT
Open ChatGPT Plugins, register a custom MCP Tunnel using the same Tunnel ID, and authorize the connection. Refer to the OpenAI connection guide for current account-specific steps.
In ChatGPT, select the plugin and call the read-only browser_status tool. Confirm cdp_ready=true, then create an about:blank tab with new_page, inspect it with take_snapshot, and close only that tab with close_page.
2.4. Troubleshooting
If the Tunnel is not connected, verify the saved ID, key, network, and local service status. If ChatGPT cannot call the tools despite a connected Tunnel, verify that its plugin is registered and selected. Do not share credentials or local logs containing private data.
3. Development
You can skip this section if you're using an installer. To work with the source, install Git, npm, and a Node.js version supported by package.json:
git clone https://github.com/ThinkerQAQ/remote-browser-mcp.git
cd remote-browser-mcp
npm ci --no-audit --no-fund
npm run check
node scripts/client-smoke.mjs
npm run audit:highcheck runs unit and compatibility tests; client-smoke.mjs checks MCP initialization and tool discovery without requiring a browser. npm run proxy-smoke verifies a real browser connection with a dedicated test tab. On Windows CMD, run npm.cmd directly.
Build the client with Go (cd client && go build -o ../bin/remotebrowsermcp .); run go test ./... and go vet ./... from client/.
Changes go through pull-request review and Squash Merge. See Installer CI and Release CI for the package builds.
4. License
Original project code is licensed under Apache-2.0. Third-party licenses are listed in THIRD_PARTY_NOTICES.md.
Browser access uses the existing proxy policy: existing tabs and arbitrary JavaScript stay disabled by default. MCP requests and results may be sent to the connected AI service. Runtime keys are kept in runtime/credentials.json, restricted to the current OS account; they are not encrypted by this client. Keep config.toml, runtime/, keys and logs private.
Please report security issues privately through GitHub Security Advisories.
Tool Schema Changelog
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
This server cannot be deployed
Maintenance
Related MCP Connectors
Let AI agents use any website you're logged in to. Sessions stay encrypted on your machine.
Let AI agents observe and control Chrome tabs you explicitly share via Browse Bridge.
Connect AI assistants to your approved computer for files, commands, browser and desktop tasks.
- DNS-DATA RemoteDeprecatedcom.dns-data
Connect AI assistants to authorized Windows and macOS devices for files, commands, and management.
Related MCP Servers
- AlicenseNot gradedqualityDmaintenanceEnables AI agents to securely control a user's existing signed-in Chrome browser through isolated tab groups, with strict per-session ownership and no cookie or token exposure.6 npmISC
- AlicenseNot gradedqualityBmaintenanceEnables ChatGPT to securely control an existing Microsoft Edge session on Windows via OAuth-protected HTTPS, including tab management, navigation, page interaction, and file upload/download through an MCP proxy.MIT
- AlicenseNot gradedqualityBmaintenanceEnables MCP clients like Codex, Claude Code, and Cursor to connect to Microsoft Edge through automatic browser discovery and the official Chrome DevTools MCP server, preserving cookies, sessions, and extensions.11 npmMIT
- AlicenseNot gradedqualityBmaintenanceEnables AI assistants to control and inspect the user's actual Chrome browser via MCP, preserving login sessions and isolating agent tasks in background tab groups without disrupting the user's workflow.MIT