About html.surf
A knowledge platform built on one idea: HTML is the right format for knowledge. Not Markdown converted to HTML. Not JSON fetched into a React component. Real, linkable, traversable HTML — the way the web was designed to work.
Why HTML?
Every knowledge tool eventually asks how to expose its data to AI. The answer is almost always: add an API, write a parser, build a connector. HTML doesn’t need any of that. It’s already structured. Links are already relationships. Headings are already hierarchy. The semantic web people were right — they just built the wrong tools.
html.surf stores your pages as real HTML and serves them as real HTML. A human reads them in a browser. An AI traverses them via MCP. A curl command can fetch them. Nothing in between. No conversion layer, no lock-in, no proprietary format to migrate away from later.
The graph
Pages link to each other. Every outbound link is tracked. Every backlink is surfaced in the sidebar. You build a knowledge graph just by writing — no ontology required, no schema to maintain. Structure emerges from content.
The navigator agent traverses this graph to answer questions. It follows links, reads pages, and reasons across the graph the same way a human researcher would — starting from one page, following threads, building context. It’s not RAG over a flat vector store. It’s a graph walk.
MCP-native
html.surf exposes your knowledge via the Model Context Protocol out of the box. One config entry and any AI assistant — Claude, Cursor, any MCP client — can read and write your pages. Your knowledge graph becomes a tool your agents can actually use, not just a corpus they can occasionally query.
17 MCP tools — read, write, search, link traversal, sharing, permissions, themes. A full read-write interface, not a read-only export.
Built for now
Most knowledge tools were built before AI agents existed. They store data in formats that require extraction, parsing, and transformation before an agent can use them. html.surf was built with AI-native access as a first-class requirement — not bolted on after the fact.
At the same time, it’s still just a website. No AI required. Write pages, link them, share them. The hypermedia fundamentals are solid enough to stand on their own.
Principles
- HTML is the format — not a source that compiles to HTML. The content is the HTML.
- Links are data — every link is tracked. The graph is a first-class citizen.
- AI is a reader — agents access knowledge by traversing the graph, not calling a custom API.
- No lock-in — export as HTML, Markdown, or DOCX any time. It was always yours.