The Model Context Protocol (MCP) represents a significant advancement in AI assistant capabilities, enabling seamless integration with external tools and services. However, as MCP adoption grows, so do the security concerns surrounding credential management.
The authorization services and OAuth flows may solve the problem for many HTTP-based MCP servers, but little exists for local services. I like the idea of having a suite of local tools and resources I can plug into an AI.
Password Managers are a well-established pattern for credential management. I would love to see MCP provide a way to plug in password providers similar to how Apple has done in iOS through their credential provider extension system. This would give users control over which passwords are passed to which tool and also prevent the AI from having direct access to passwords. If anyone is working on such a thing, I’d love to hear about it.
For my personal use, until there is a generic solution to this problem, I’m creating simple MCP tools that integrate with 1Password directly to keep my passwords out of the hands of AI and off the filesystem. For example, I put together one recently to access MySQL Databases using credentials from 1Password.