Make it easier for AI agents to consume the docs.

[webrgp]

I've been using Context7 MCP to add CraftCMS docs as context for my prompts when working on new features or fixing bugs for client projects. It works mostly well but here are some ideas to make it even better:

Reducing Token size

At of today the Craft CMS docs clocks at 1.5M tokens. This is mostly due to the fact that we have one repo with all versions for Craft and Commerce. It would be nice feed the agent just the project relevant version of the docs.

There is a couple ways to achieve that, with different degrees of complexity.

We could start by adding a context7.json to the root of the docs to fine-tune exclusions to reduce the total token size.

Alternatively, we could skip Context7 altogether and instead have the docs build step generate llms.txt for the different versions of the docs that we can point agents directly to it.

[webrgp]

Here's an example from the Model Context Protocol (MCP) site.

The MCP site has a special endpoint similar to a sitemap file but used for LLMs:
https://modelcontextprotocol.io/llms.txt

Additionally, there is another endpoint that provides the entire documentation (plain text format) in one place for LLMs:
https://modelcontextprotocol.io/llms-full.txt

This very convenient in this age of AI.