llms.txt vs MCP

llms.txt is a file a model reads. MCP is a protocol a model calls. One describes a site, the other answers questions about it.

llms.txt is a file a model reads. MCP is a protocol a model calls. One describes a site in markdown, the other exposes tools that answer questions and return live data.

llms.txtMCP
Shapea static markdown file at a fixed patha server the client connects to
Directionthe model reads and moves onthe model calls a tool and gets an answer back
Freshnesswhatever the file sayswhatever the tool computes at call time
Authenticationnone. It is public.a key, so an answer can be scoped to one account
Good fortelling an agent what your site is and where to readletting an agent ask a question you cannot precompute

Both, not either

They are not competitors and a site can publish both, which is what Domduck does. /llms.txt says what the product is and links the clean markdown copies of the documentation. The MCP server answers questions the file cannot: the rating series for one domain, the crawler policy for another, what changed last week.

The dividing line is whether the answer exists before the question. A page of documentation exists already, so a file is the right delivery. A time series for a domain nobody asked about yet does not exist as a page, so a tool is.

What Domduck records

Neither of these is a measurement about your domain. Domduck records whether your site serves an llms.txt, and it does not detect whether you run an MCP server, because there is no fixed path to probe for and guessing would produce a false negative on every site that runs one somewhere else.

Why it matters

If you are choosing where to spend an hour: publish llms.txt if your audience points agents at your site, and build an MCP server if your product holds data somebody would ask a question about. A documentation site wants the first. A database wants the second.

Primary source: modelcontextprotocol.io