Build on it

Everything a reader sees here, a program can read too, with nothing to sign up for: the concepts, the learning paths, the certification roadmaps, the news, the resources, and the state of the platform — what is in preview, what was renamed, what is ending. One MCP server, the same operations over REST, and the vault as plain files.

Connect an assistant

The MCP endpoint is https://lakenaut.dev/api/v1/mcp. No key, no OAuth: paste the URL and the tools below appear. Then ask your assistant something it would otherwise guess at — "what is the difference between a streaming table and a materialized view on Databricks?", "which concepts does the Data Engineer Associate exam weigh most?", "what is in public preview right now?" — and it answers from the pages here, with the link to each one.

  1. Claude.ai: Settings → Connectors → Add custom connector. Name it Lakenaut, paste the URL, leave authentication empty. In a chat, switch the connector on from the tools menu.
  2. Claude Code:claude mcp add --transport http lakenaut https://lakenaut.dev/api/v1/mcp
  3. ChatGPT: Settings → Connectors → Create (developer mode), with the URL and no authentication.
  4. Cursor, Windsurf, and most other clients take the same thing as JSON:
{
  "mcpServers": {
    "lakenaut": {
      "url": "https://lakenaut.dev/api/v1/mcp"
    }
  }
}

Streamable HTTP, stateless, JSON responses. A client that still speaks only the SSE transport will not connect; everything released since mid-2025 does.

What you can call

13 tools
whoamiGET /api/v1/Who this caller is, what it may do, and where to start.
taxonomyGET /api/v1/taxonomyThe site's vocabulary: areas, paths, roadmaps and their domains, concepts, and the closed list of topics.
concepts_listGET /api/v1/conceptsEvery concept: id, title, summary, area, level, status, updated, certifications, and its path in the vault.
concept_getGET /api/v1/concepts/:idOne concept: its frontmatter, its markdown body, and what the site knows around it (maturity, renames, backlinks, paths, exam domains).
roadmaps_listGET /api/v1/roadmapsThe certification roadmaps with the exam guide each follows, its version, and the concepts per domain.
vault_listGET /api/v1/vault/filesList every file of the vault with its etag, size and time of the last write.
vault_readGET /api/v1/vault/fileRead one vault file.
vault_versionGET /api/v1/vault/versionThe current version of the vault: changes whenever any file does.
news_listGET /api/v1/newsNews items, newest first.
news_getGET /api/v1/news/:idOne news item.
resources_listGET /api/v1/resourcesResources, filtered by concept, area, type or state.
platform_getGET /api/v1/platformThe radar (features in preview), the renames and the end-of-support dates, as the site shows them now.
hints_listGET /api/v1/hintsExam hints, by certification, concept or status.

Start with taxonomy for the ids, then concept_get for a page. A concept comes back as its frontmatter, its markdown, and what the site knows around it: maturity, old names, backlinks, the paths and exam domains it sits in.

The same over REST

openapi.json 
curl https://lakenaut.dev/api/v1/concepts/auto-loader
curl "https://lakenaut.dev/api/v1/news?since=2026-09-01"
curl https://lakenaut.dev/api/v1/platform

GET takes the query string; the answer is JSON, never cached. The OpenAPI document describes every operation and imports as tools into most agent frameworks. CORS is open, so a page on another site can call this directly.

Or just the files

llms.txtAn index a model can read: every page with a line about it.llms-full.txtEvery published concept in one file, with its links, documentation and resources.The vaultThe published pages as markdown with wikilinks, packed nightly. Opens in Obsidian with no plugins.JSON/taxonomy.json, /concepts.json, /graph.json, /roadmaps.json, /preview.json, /lifecycle.json, /search.json — and /rss.xml for the news.

Limits, and what stays closed

Sixty calls a minute per address without a token, which is plenty for an assistant and not enough to mirror the site in a loop. If you are building something that needs more, ask: a token with the same reader's view has no limit, and a token is also the only way to anything else. Drafts, the review queue, the agent's jobs and every operation that writes stay behind scopes issued by hand.

Nothing here writes. The agent that keeps the site proposes changes a person accepts; a connector pointed at this endpoint can read and cannot propose. Readers report a problem or suggest a resource from the page itself.

The documentation is the authority. Every concept lists the pages it was written from in sources. Facts about how the product behaves come from those; this site says how to understand them.

Licence

The writing — concepts, areas, paths, roadmaps, practices — isCC BY-NC-SA 4.0: use it, quote it, build on it, with a link to the page it came from (https://lakenaut.dev/concepts/<id>/), under the same licence, and not to sell. The code in every fenced block is MIT. Facts about a product are nobody's.

Lakenaut is an independent project, not affiliated with Databricks; product names are used as nomenclature only. The vault download carries the full text of both licences.