Replies: 3 comments 2 replies
-
Came here searching for just this post. Yeah I think this would make a great match for Avante. Will keep my eyes on this closely! For some additional context, here's their intro blog post about it: https://www.anthropic.com/news/model-context-protocol:
|
Beta Was this translation helpful? Give feedback.
-
So if I understand correctly basically what this does is expose a set of tools on an api endpoint that other applications can integrate with and include in prompts sent to the LLM? If that's true then what I understand the architecture would be to have avante initialize/discover mcp servers and then make sure those tools are included when it sends avante ask and/or completion requests to the LLM. I imagine that would require an ecosystem of mcp servers to be built out to handle a variety of tasks (local code search, documentation search, etc). From a installation/maintenance standpoint also not sure what the best path would be. The reference implementations are python/typescript but the protocol is open. Should the decision be made to support it, it would probably make more sense to do the mcp brokering in lua and have a way for mason to install mcp servers similar to what it does for lsps/daps/etc. |
Beta Was this translation helpful? Give feedback.
-
I think MCP can be an extensions of tool use. At the end of the day they are built on top of how these models return structured outputs. |
Beta Was this translation helpful? Give feedback.
-
https://modelcontextprotocol.io/quickstart
Anyone hear about MCP protocol proposed by AnthropicAI? Can we provide a server like this to enhance experience?
Beta Was this translation helpful? Give feedback.
All reactions