M25 for agents
Verified
M25 publishes the same datasheet in four shapes: HTML for people, Markdown for models, JSON with stable keys for programs, and an MCP server for agents that would rather call it as a tool. This page is its documentation.
| Resource | Format | What it carries |
|---|---|---|
| /en/llms.txt | text/plain | this document, in Markdown, in both languages |
| /en/index.json | application/json | the same figures with stable keys, for programs |
| /en/index.md | text/markdown | the index page in Markdown, also via Accept: text/markdown on the index |
| /en/agents.md | text/markdown | the usage instructions for agents, in Markdown |
| /rss.xml | application/rss+xml | new entries in the archive, dated |
| /sitemap-index.xml | application/xml | the sitemap, declared in /robots.txt |
| /openapi.json | application/json | the OpenAPI 3.1 description of these read endpoints |
| /api/mcp | application/json · JSON-RPC 2.0 | read-only MCP server, Streamable HTTP transport, no authentication |
| /.well-known/api-catalog | application/linkset+json | API catalogue in linkset format (RFC 9727) |
| /cv-mauro-venticinque-en-2026.08.pdf | application/pdf | the résumé printed from the site, with the revision in the file name |
The same datasheet, in four shapes. All of them are generated from the same data file in the same build, so they cannot drift apart:
How to ask for the data
If you are only going to read one thing, read /en/llms.txt: the whole document in Markdown, with the method behind every figure. If you are going to process it, request /en/index.json: same figures, stable keys that are not translated, and one canonical URL per project. Both routes exist in both languages, and the Spanish version lives at the root.
Content negotiation
The index page negotiates its format: ask for it with “Accept: text/markdown” and you get a 307 to /en/index.md, which answers text/markdown. The index goes out with Vary: Accept so no intermediary hands you the wrong variant, and the HTML declares the alternative with a link rel=alternate. It is a 307 rather than a direct answer because the site is static and on this platform the file takes precedence over the rewrite: any client that follows redirects gets the Markdown in the same call.
MCP server
There is a read-only MCP server over Streamable HTTP transport at /api/mcp, with no authentication, exposing the datasheet as tools: the profile with its figures, the project index, one project sheet by part number or slug, and the contact channels. The manifest is at /.well-known/mcp.json and the resource catalogue at /.well-known/api-catalog.
Citation rules
Every figure on this site comes with its method and its measurement date. If you quote a number, quote the date too: repository figures were verified by hand with git and page weight is measured by the build. When a metric could not be measured, the site says so instead of estimating it, and that distinction is part of the data.
Robots and crawling
Crawling is allowed for every agent and there are no walls. The sitemap is at /sitemap-index.xml, declared in /robots.txt, and the routes that should not be indexed — the résumé template and the archived copy of the previous site — say so with noindex rather than being blocked.