The GitHub MCP Registry just landed 🙌
Whether you’re building with GitHub Copilot, agents, or any AI tool that speaks MCP, this is the place to find what you need. github.blog/ai-and-ml/gith…
I was just browsing this last night or at least I think it was the same thing. Just not the blog post. I hope that this isn’t just a perpetual list of alphabetical MCP’s.
We need an MCP and agent orchestration just to browse explore test and decide which MCP is to use, it’s a competitive space and anyone can build a MCP server so please, let’s find a way to enable the community to make this experience better. I wanna bring solutions just as much as I bring critique, but there are so many ways to go about it and so much more deep of a pocket than I have for those that wanna monetize the traffic to their MCP landing page.
Let that last line sink in their NCP landing page, which maybe if you think three more times you’ll find out that you need to pay wall or an API key or it’s just a way to get traffic from the well built, likely by vibe, landing page capitalizing on “MCP”.
Any @code or @github DevRels out there want to help improve this MCP DX please?
@github A question: How can I direct to the github.com/mcp from my Github home page? I didn't find any shortcut to do that. If this can be done, it will be more convenient for people to find the mcp plugins.
@github This is the real key to agentic workflows.
Generic LLMs are impressive interns. But specialized tools, exposed through a protocol like MCP, are senior engineers.
Connecting the intern to the senior engineers is how you get real work done.
MCP kinda sucks though. We should just be using json schemas and including return types. The ai-sdk has a better setup with their tool definitions.
I've been testing graphql instead and it works really well. I had to add a discovery method for finding tools beyond a simple introspection query but it seems to be a clearly better option than using MCP or most tool setups directly - assuming our tools have well defined inputs and outputs.
Many tool setups, MCP or not, can bloat the LLM with excessive and largely irrelevant tools. Additionally, I believe (although I need to verify this) graphql queries use less tokens than json, and the LLMs can make many queries/tool calls at once. Maybe smaller models might struggle with graphql(I doubt it) but all current flagship models are more than capable of using it without an issue.
ai-sdk.dev/docs/reference…
@github This is a game-changer for devs! Finally, a super easy way to find all the MCP servers in one place. Excited to see how this powers up new AI tools and workflows!