The elephant in the MCP usability room

Here it is: 🐘

But seriously, MCP is a great protocol to connect AI and data. Unfortunately, the usability of this protocol is a massive challenge for non-technical users. The majority of servers require a comprehensive multi-step API key extraction process if you want to run them locally and not share your sensitive data with a third-party MCP server hosting platform.

For example, the amazing google_workspace_mcp (611 GitHub stars at the moment of writing this doc) asks its users to "paste your Google OAuth credentials," which sounds like something you wouldn't ask a non-technical person to do. Even for a technical person, it takes about 20 minutes to register in Google Cloud, create an application there, and put the file in the right place.

We believe that:

1. End-users should be able to use MCP without messing with API keys.
2. Archestra should work with existing open source MCP servers (not rewriting or running in a third-party cloud like other providers do)

In order to support any open source MCP server and provide the best UX to end-users, we developed a few options for authenticating users in Archestra and the subsequent MCP server.

Option One: Updating Your MCP Server

If you own the MCP server code, that's great! To make it work in Archestra, we ask you to comply with the MCP specification's authorization flow, which was introduced in version 2025-06-18.

(This section is under construction and will be updated, issue for this)

Option Two: Configuring Archestra to Support Third-Party MCP

For existing MCP servers that cannot be modified, Archestra provides two authentication methods:

Browser-Based Authentication

Archestra includes a built-in browser that can automatically extract authentication tokens from web services.

See Browser-Based Authentication for implementation details.

OAuth Integration

For services that support OAuth 2.0, Archestra provides a standardized OAuth flow with PKCE security.

See OAuth Providers for setup instructions.