[ For agents ]

The MCP endpoint

Search and read the @sdxc package documentation over the Model Context Protocol: one stateless endpoint, every tool read-only, no credential.

https://sdxc.sergiodxa.com/mcp is a Model Context Protocol server over the 87 packages documented on this site. It speaks stateless Streamable HTTP, so there is no handshake and no session to keep: one POST is one exchange.

Nothing here is behind a credential, every tool is read-only, and everything a tool answers with is already public as a page.

Connecting

Point a client at the endpoint. Most take a URL directly; the ones that read a configuration file take this:

mcp.json

{
	"mcpServers": {
		"sdxc": {
			"url": "https://sdxc.sergiodxa.com/mcp"
		}
	}
}

What it offers a model

Every page on this site already answers on a .md URL, so anything that can fetch can already read all of it. What it cannot do is find the thing, and that is what these tools are for.

Tool	What it answers
`search_packages`	"Is there a package for X?", over every name, description and README
`list_packages`	The whole set, grouped by the problem each group solves
`get_package`	One package in full: its README, its exports, and what installs alongside it
`search_docs`	Where something is explained, answered with the heading's own URL

What it offers a reader

Every guide and every package reference is also a resource, listed by name, so a client's picker carries the whole corpus and a person can attach one page before they start. Each resource's URI is the page's own markdown address, which means attaching it and fetching it give the same text.

Reading it another way

The same content is published as a markdown map of the site, as a search index, and as a feed.