MCP Router

Connect multiple Webflow workspaces to your AI tools.

A local desktop app for macOS and Windows by Kvalifik. Give your AI tools access to projects across multiple Webflow workspaces at the same time through one MCP connection, with access controls for each project.

MCP Router is independent software. It is not affiliated with, endorsed by, sponsored by, or approved by Webflow, OpenAI, Anthropic, or the other services it connects to. Product names identify compatible services only.

Manage connections and project access in one place. All project and connection names in these screenshots are fictional demo data.

Install on macOS

Choose mac-arm64 for Apple Silicon (M1 or later), or mac-x64 for Intel Mac.

1. Open the GitHub Releases page and download the latest MCP-Router-…-mac-arm64.zip or MCP-Router-…-mac-x64.zip. Choose the app ZIP, not GitHub’s “Source code” ZIP.

2. Extract the ZIP and drag MCP Router.app into Applications.

3. Open MCP Router. This release is ad-hoc signed, without an Apple Developer ID or notarization. If macOS blocks it, first attempt to open it, then go to System Settings → Privacy & Security → Open Anyway and confirm opening the app. Only do this for a download you trust; don’t disable Gatekeeper. Managed company Macs may require IT approval. If macOS says the app is damaged, download a fresh copy and report the release version if it persists.

4. Choose Add connection, authorize Webflow, then enable the projects and permissions you want to use. New projects start disabled.

5. Open Apps, connect your AI app, and follow its restart/reload instructions.

Related MCP server: Gealo MCP Server

Install on Windows

1. Download MCP-Router-…-windows-x64.zip from the GitHub Releases page when available.

2. Extract the entire ZIP into a permanent folder, such as %LOCALAPPDATA%\Programs\MCP Router.

3. Run MCP Router.exe. Keep its supporting files and resources folder together. This build is unsigned; Windows may display an unknown-publisher warning.

4. Add a connection, authorize Webflow, enable your projects, then connect your AI apps.

Windows support is new and should be smoke-tested before broad distribution. App discovery covers common per-user and system installations; ChatGPT/Codex detection includes Microsoft Store installations with a bundled CLI, standalone native CLI installations, and npm CLI installations. For custom installation locations, use the copied MCP configuration. Quit through the MCP Router menu before replacing the app folder during updates.

Updating

When a new public GitHub release is available, the app shows Download update. You can also use Settings → Check for updates. Downloading opens the release page; the app does not install updates or restart automatically.

To install an update: finish active work, save any settings, quit MCP Router, and replace the app in Applications with the newly downloaded copy. Reopen it and reload/restart connected AI apps. Connections and settings stay in the separate application-data folder. Do not delete that folder when updating.

Update checks use public releases from Kvalifik/mcp-router.

What it does

• Work across multiple Webflow workspaces without switching connections in your AI tools.

• Authorize Webflow access with OAuth, without sharing account passwords.

• Discover projects automatically; new projects start disabled.

• Control read, write, delete, and publish actions per project.

• Use read-only, no-publishing, full-access, or custom permission presets.

• Pin frequently used projects and organize connections.

• Configure local MCP clients, including Codex, Claude Desktop, Claude Code, Cursor, VS Code, and Gemini CLI.

• Follow system appearance or choose light or dark mode.

The app also includes separately authorized Beta MCP support. Beta access requires separate authorization and remains subject to Webflow’s applicable terms and availability.

Status

Build targets are macOS on Apple Silicon and Intel, and Windows x64. Windows builds are unsigned; Mac builds are ad-hoc signed without an Apple Developer ID and are unnotarized. See SECURITY.md for the security model and known limitations.

Build and run

Requires Node.js 22 or later and npm. Build Mac releases on macOS and Windows releases on Windows.

npm ci
npm run build:ui
npm run desktop

Run npm test for the automated suite and npm run build:mac to create dist/MCP Router-darwin-arm64/MCP Router.app.

For Intel Mac, run BUILD_ARCH=x64 npm run build:desktop. For Windows, use PowerShell:

$env:BUILD_PLATFORM = "win32"
$env:BUILD_ARCH = "x64"
npm run build:desktop
python scripts/package-release.py

On Mac, run BUILD_ARCH=arm64 python3 scripts/package-release.py (or x64) after building to validate and create the release ZIP. The release workflow builds all three targets, runs tests on each host, and combines their checksums before creating an optional draft.

The app stores desktop state in ~/Library/Application Support/MCPRouter/router/ on Mac and %APPDATA%\MCPRouter\router\ on Windows. Existing desktop state is moved automatically from the former directory on first launch. The bundle identifier is dk.kvalifik.mcp-router; AI clients use mcp_router_for_webflow. Legacy client entries are migrated with settings preserved. Running npm start directly uses .local/ unless ROUTER_DATA_DIR is set. Do not run two instances on port 43127.

The desktop app signs into its local dashboard automatically. For npm start browser access, open the single-use URL in dashboard-login.txt inside the state directory. It expires after ten minutes; restart to generate a new link. Treat that file as private. Direct unauthenticated dashboard requests are denied.

Set up

1. Open the app and choose Add connection. Review how authorization and router permissions work, then choose Continue to Webflow.

2. Authorize the intended sites/workspaces with your own Webflow account.

3. Review permissions and enable the projects you want your AI client to access.

4. Choose your client in Apps, then restart or reload it as instructed.

A configured checkmark means configuration was saved, not that a running AI session has loaded it. After connecting, the checkmark stays visible until the next hover or keyboard focus. Hover or focus the configured app again to show its reconnect icon. The ChatGPT/Codex entry configures a detected local Codex MCP client; it does not add a connector to the ChatGPT website.

You can authorize fewer projects or permissions. Router permissions only limit access already granted by Webflow; the router cannot inspect the full Webflow permission grant. Review authorized projects under More options → MCP version.

Moving the installed app changes the executable path. Reconnect each client after moving or renaming its bundle. Other registrations, project overrides, and client-specific restrictions remain relevant.

Permissions and limitations

Choose a permission preset or control individual actions for each project.

Router permissions apply only to calls passing through MCP Router. They do not reduce permissions in Webflow itself or in separate integrations. The app cannot prevent the same OS user or an agent with local filesystem access from changing its code or stored settings.

New projects are disabled but initially have all known permissions selected; review the defaults before enabling them. No publishing blocks explicit publish/unpublish operations. Some writes can affect live settings without a publish call. A sent write cannot be rolled back by disabling the project.

Each user needs their own provider authorization. OAuth scopes, account roles, plan limits, and Designer session requirements still apply. Stable and Beta have separate grants; missing Beta access does not fall back to Stable.

Read the usage guide, privacy notice, security policy, and usage terms.

License

Source-available under the MCP Router Source-Available License 1.1. Business use and paid client work are allowed. You may modify the app and redistribute meaningfully changed versions for free (see LICENSE section 4a for exceptions). Selling the router, charging for access, and offering its core functionality as a paid service require separate permission.

This is a custom source-available license, not an OSI-approved open-source license. The full license controls; see licensing notes. Third-party components retain their own licenses.

Contributing and support

See CONTRIBUTING.md. Contact Kvalifik ApS at contact@kvalifik.dk. Please report security issues privately rather than attaching credentials or client data to GitHub issues.

Third-party notices · Branding