modelcontextprotocol/servers ? reverse-engineered prompt

I want to build a project that lets me give my Large Language Model (LLM) access to a bunch of different tools. Think of it as a collection of small servers that the AI can call to perform specific tasks in the real world.

For the initial version, let's include a few key tools. I'd like one that can securely access and manage files on the local filesystem, another that can interact with a Git repository to read code, and a simple one for fetching content from websites. It would also be great to include a "memory" tool that gives the AI a persistent knowledge base to use across conversations.

The whole project should be built in TypeScript and structured so that it's easy to add new tool servers later on. I've seen examples of integrations for things like Slack, Google Drive, and databases, and I want my system to be extensible like that. You can look at how the "Model Context Protocol" works for inspiration on the best architecture.