Skeleton MCP for Docs #4269
Description
Describe the feature in detail (code, mocks, or screenshots encouraged)
Currently Skeleton only has llm files for language models. While these are great there currently is not a great way to have agents access them and use them consistently. My current understanding of the options are:
- Link to the llm file in an agent.md or equivalent usually does not work as most agents and models will not go out of their way to make a remote fetch to a url unless specifically prompted to do so.
- Download the appropriate llm file and include it in your repo so the agent always has access. However this can quickly have outdated information and depends on the given models strengths at searching large files for information.
MCPs on the other hand are a standard that agents are actively trained to use today and can use them well. They are willing to make the tool calls to use them. In addition an MCP would have access to up-to-date documentation.
Provide relevant links, wireframes, prototypes, or additional information.
An example of this is the Svelte MCP. Before I used it I regularly would have to tell my agents to use runes and the Svelte 5 syntax as many were trained before runes were the standard. Since using the Svelte MCP it has up-to-date information and uses it so I never have to tell it to use runes.
Skeleton would benefit in the same way, as often agents see tailwind as a package and assume to do it the tailwind way, agents wouldn't need redirection to do things like text-primary-600-400 instead of text-primary-600 dark:text-primary-400 and would just do it properly the first time.