haha I like the word vibefrastructure!

Yes thank you for sharing your thoughts. The MCP interfaces definitely sound promising. However I think it does not stop people from building other MCP servers . And different people may have different opinions regarding how to design this interface for a search MCP server.

TBH I think all of these problems are still very new (remember MCP is only 6 months old) and I think we need to wait and see how things evolve.

Hopefully, in the end we will all converge to the real standards. But I understand your concerns now and I agree that right now it is a bit messy.

Yes I agree with you. What are the major shortcomings that you can think of right now (especially the ones that we have not solved)?

I think the two biggest issues are probably:

1. Giving the model too many choices. If you have a lot of options (like a bunch of MCP servers) what you often see in practice is that it's like a dice roll which option is chosen, even if the best choice is pretty obvious to a human. This is even tough when you just have a single branch in the prompt where the model has to choose path A or B. It's hard to get it to choose intelligently vs. randomly.

2. Global scope. The prompts related to each MCP all get mixed together in the system prompt, along with the prompting for the tool that's integrating them. They can easily be modifying each other's behavior in unpredictable ways.

Makes sense. Both are hard problems I agree.

yes right forgot to mention this.

Thanks! I think our hosted version has an API endpoint satisfy your needs on mobile devices. You can use our API to call any remote MCP server tools directly. https://docs.klavis.ai/api-reference/mcp-server/call-tool.

Regarding SDKs, yes we are planning to add SDKs as well. The Vercel AI SDK is definitely an interesting one. I will take a look.

Hi yes it sounds like Klavis AI meets your requirement. Can you shoot me an email at xiangkaiz@klavis.ai? Happy to chat!

Hi yes so if you just use it yourselves locally in Cursor, I think you can definitely just use the official GitHub MCP. But there might be more use cases where people want to connect MCP servers in their AI agents and deploy it in the cloud, then the local GitHub server would not be a fit.

The UX for end user OAuth is the standard GitHub Oauth flow. You will see a webpage saying that an GitHub Oauth App is requesting access to your github and you need to select yes/no.

Yes thank you for the good feedback. We will update our doc to reflect this.

> how would I add this to my agent? Would you have One MCP To Rule Them All?

If you are talking about the hosted version, you can go to our website to grab an API key and call the corresponding endpoint. https://docs.klavis.ai/documentation/mcp-server/github is the doc for GitHub MCP server.

in terms of one mcp to rule them all, it is something on our roadmap now. We are thinking that you can connect to all of the MCP servers we provide with a single URL and you can turn on/off the MCP servers in a simple UI or through API call. Let me know if this is what you have in mind.

Right now, as you know we’re in a total Wild West with this. So the experience is different for each client (VS code, cursor, etc)

To me, the ideal the UX is similar to the extensions UXs in each IDE, you search, click install, and get on with your life after doing an OAuth dance. Or even a discovery mechanism built into the agent itself. Sort of a “have you considered installing this? I can talk directly to JIRA for you” etc

The next best thing would be a catalog where I search, and with a single install button that would add the MCP to a certain IDE. But do the IDEs make this sort of API available, given that they all want to differentiate?

Yes thank you! the newest MCP spec added the authentication part but it seems that people think it is still not perfect and are doing more modifications to the auth part. E.g. https://github.com/modelcontextprotocol/modelcontextprotocol.... We will also keep an eye on the spec development.