Agent-friendly user docs for Wagtail The Wagtail project published its user guide documentation in four agent-oriented formats — an MCP server, Markdown served via content negotiation built on Wagtail 8.1's Vary HTTP header support, a revised llms.txt sitemap, and experimental WebMCP — after testing them across 33 information retrieval tasks with 5 harnesses and more than 10 models. The project reported that agents used llms.txt in only 11% of runs when the file was linked from Markdown responses as a "documentation index," versus below 1% when it was unlinked, rendered as vanilla HTML, or inlined in every response, and 52% when the agent's search functionality was unavailable. The project called the formats "clearly not a transformational technology" but worth setting up for added redundancy and ease of use for agents. Agent-friendly user docs for Wagtail Four new documentation formats for AI agents After weeks of testing, we are now making Wagtail’s user guide documentation available in four new formats https://guide.wagtail.org/en/about/documentation-formats/ . Try them out and let us know what you think The new formats Here they are in the order of "most" to "least" useful based on the evidence we collected. MCP server Once set up, the user guide MCP server https://guide.wagtail.org/en/about/documentation-formats/ mcp-server allows direct fetching of docs, and site-wide search, in a format optimized for agents doing tasks. Markdown with content negotiation Every page of the guide now serves Markdown variants on request, based on the Wagtail 8.1 https://wagtail.org/blog/wagtail-8-1/ support for the Vary HTTP header and Markdown conversion. This is useful for task agents but also crawlers for training and search. llms.txt We revised the guide website’s llms.txt https://guide.wagtail.org/llms.txt file to facilitate navigation, using it as an agent-friendly sitemap. While almost no agents reach for this file, it works very well when signposted for them, particularly for simpler agents with no built-in search tools. WebMCP Just like the MCP server, you can use webMCP https://developer.chrome.com/docs/ai/webmcp to search and retrieve pages of the site. This is very experimental so only works with a limited number of browser-based agents. Discovery You can find all those options and more in the site’s Documentation formats https://guide.wagtail.org/en/about/documentation-formats/ page. For agents, the site also advertises the formats according to the Agentic Resource Discovery specification https://agenticresourcediscovery.org/ . How we tested them We built an evaluation suite of information retrieval tasks, for common documentation requests. It compares 5 harnesses and 10+ models on 33 tasks, which we ran over all of the different formats to increase the success of the tasks as well as the cost / efficiency. For example, for the runs where agents used llms.txt, usage was: - Below 1% when not linked from anywhere - Below 1% when rendering vanilla HTML for agents - Below 1% in Markdown but with llms.txt contents inlined in every response. - At 11% when the file was linked from the Markdown responses as a "documentation index" - At 52% when the agent’s search functionality wasn’t available. This is clearly not a transformational technology but it’s well worth setting up for the added redundancy and ease of use for agents. If you want to see our full results and how this applies to the wider Wagtail ecosystem, be sure to join Agent readiness for Wagtail sites https://wagtail.org/wagtail-space-2026/agent-readiness-for-wagtail-sites/ as part of Wagtail Space 2026 https://wagtail.org/wagtail-space-2026/ later today And we also have a corresponding Guide to GEO presentation aka really good SEO https://wagtail.org/wagtail-space-2026/a-wagtail-editor-s-guide-to-geo-aka-really-good-seo/ for editors.