As soon as I saw the MCP announcement - when dsp annoucned they had open sourced the repo - and started working on my own server [1]...it's quite interesting that anthropic has decided to open source this protocol. I have a couple of questions and if anyone from anthropic reads this please also let us know you're from anthropic.

1/ Why did you built a protocol like this without getting other LLM providers on board - because I believe that would have helped promoting and making this more of a unified effort, now it feels like Anthropic is running the show no?

2/ why is there a need to cenralize this now? surely anyone knows how to use npm or uvx - why do we need a directory? usually this means someone will be moderating this directory, but doesn't that defeat the purpose of letting people contribute (i.e. you will probably reach some bottleneck around validating each server...imagine if npm or pypi had to validate everything...)

3/ What are the plans for MCP long-term?

4/ Side-question: why isn't Anthropic pushing more on making it simple standardizing tools - i.e. openAI has Custom GPTs and they're basically a big RAG play but I can see how slowly these custom GPTs will be able to access more and more external services - what's the Anthropic equivalent?

Thanks for any partial answer, I'm on the MCP official discord too if anyone wants to geek out :)

[1] https://github.com/toolhouse-community/mcp-server-toolhouse

This directory is not affiliated with Anthropic. As much as I am pleased by the idea that someone could somehow confuse my hobby project with something that Anthropic could ship, what you see is a hobbyist effort to help curate the community resources. I just happened to have been involved since the early announcement and it kinda snowballed from there.

As to 'why', I touch on this in the original comment under the post, but:

* it helps with discoverability by standardizing how servers are described

* it helps with security by automating audits (non-trivial effort) and disseminating security alerts

In terms of scalability, the current setup, while not perfect, has been holding the fort. I've built lots of tooling around virtualizing environments for testing MCP servers, tools for introspection, auto-setup using LLMs, etc. Will see what the future holds, but I am ready to see this grow to thousands of servers.

For these other projects to succeed, they need to figure out how to create value beyond listing servers, e.g. I am doing this now by testing servers in sandboxed environments, running dependency security audits, etc. However, this isn't cheap. So whichever the projects I am competing with also have to have a financial backbone. In my case, I’m building a service—VMs for AI personal assistants—that will integrate this directory. Users will be able to one-click add any MCP to their Glama workspace, etc. Therefore, it made sense to invest into all of this infrastructure. Meanwhile, giving this to the community builds trust and interest in what I do. I see this as a win-win.