For twenty years, documentation had one audience: people. You wrote for developers skimming at 2 a.m., for customers mid-task, for the new hire on day three. That audience is still here. But a second one has arrived, and it reads differently.
Coding agents look up your API before writing an integration. Claude and ChatGPT answer questions about your product whether or not you participate in the answer. Support copilots draft replies from whatever version of your help content they can reach. The question is no longer whether AI agents will read your knowledge. It is whether they will read the version you maintain.
Machine surfaces, published automatically
Every GraphStaff site ships its machine audience surfaces automatically. An MCP server lets agents search and read your content directly; your customers can hand it to the assistant they already use. llms.txt gives language models a canonical summary of what your site covers. And every page is available as clean Markdown, without navigation chrome or layout noise.
None of this is a separate publishing pipeline to maintain. Publish once, and the human site and the machine surfaces stay in step.
The same rules for every reader
An agent is a reader with an access level. GraphStaff applies the same permission rules to MCP queries and machine surfaces that it applies to navigation and search, so customer-only or internal content stays protected no matter who, or what, is asking.
Answers you can stand behind
When an agent answers from your maintained content, it cites real pages you control. When something is wrong, you fix the source once and every surface updates. That is a much better position than hoping a model’s training data got you right.
Curious what your docs look like to an agent? Our own llms.txt is right there. No humans allowed, but we will not tell.