Why superlore
Half your docs are read by agents now, not people. Every other tool bolts an MCP onto the old way of writing them — superlore rethinks the document itself, so one source serves a clean site to people and a typed MCP corpus to agents.
Documentation broke the day AI started writing it.
For years the thinking lived on a whiteboard — FigJam, Excalidraw, a wall and a marker — and the doc was the write-up. Then agents started shipping whole features end-to-end. Ask one for the spec and you get a flawless wall of Markdown: flat, and only as useful as you have time to read. The thinking that lived on the board flattened with it.
superlore is the fix. Author once, and the same source renders a clean, visual site for people and a typed MCP corpus for agents — not a docs page with an MCP bolted on. One corpus, two first-class readings.
Agent-native, not retrofitted
AI writes, reads, and maintains your docs now. superlore is built for that first — and still beautiful for the people who read it.
Whiteboards, inside the doc
Brainstorms live as structured data — exact nodes, edges, and relations an agent can query, not a flat image. The board is the graph.
One corpus, shared context
Every team's knowledge in one place, one MCP call from any agent. The company's shared memory — not fifteen scattered wikis.
Coming from another tool?
You don't have to throw away what works. superlore reads MDX, deploys to your own Vercel (or anywhere), and a migration agent ports your existing site page-by-page. Here's what changes — and what you keep.
From Mintlify
Keep the components you love — Cards, Steps, Tabs, and Callouts work the same. Gain a native MCP over the same content (not a separate AI add-on), typed dual-representation, and a Canvas. You self-host on your own Vercel — no per-seat SaaS.
From Obsidian
Your canvas becomes a typed graph an agent can actually query — not a private .canvas
file. Your vault's knowledge ships as a real site and an MCP, so your team and their agents
read the same corpus.
From Docusaurus / Nextra
Keep MDX-in-repo and full ownership. Drop the boilerplate: structural components — Timeline, Board, Comparison, Decision — that render for humans and serialize for agents, plus an MCP with zero extra wiring.
From GitBook / Notion
Move from a hosted silo to MDX you own in git. Same readable site, and your content becomes machine-readable knowledge any agent can pull over MCP.
superlore vs the alternatives
| Criterion | superlore | Mintlify | Docusaurus | Obsidian |
|---|---|---|---|---|
| MCP-native — agents query the same corpus | ||||
| Dual representation — typed data, not OCR | ||||
| Whiteboards as a typed graph (Canvas) | ||||
| Author once in MDX, in your repo | ||||
| Open source · you own it | ||||
| Self-host anywhere (Vercel, etc.) | ||||
| Beautiful out of the box |
This table is dual-representation — live
Every cell above is typed data, not a rendered checkmark. An agent reading this page over the
MCP can answer "which tools are partial on MCP-native?" directly from the
corpus — no OCR, no guessing. That's the whole idea, demonstrated on this very page.
Bring your docs over
Migrate an existing site
Point an agent at your Mintlify / Docusaurus / GitBook / Markdown docs — it plans, you review, it ports every page to superlore.
Start fresh in minutes
npm create superlore — or let Claude scaffold it. A KB that builds, deploys, and serves an MCP
out of the box.
In the creator's words
For years, every project started at a whiteboard — FigJam, Excalidraw, a wall and a marker. I'd plan the whole shape of a thing before a line of code; the board was the understanding.
Then Claude started shipping whole tickets end-to-end — brilliant at it. But ask for the spec and you get a flawless wall of Markdown: flat, and only as useful as you have time to read. The thinking that lived on the whiteboard flattened with it.
So I built superlore and moved the whole company in — engineering, product, roadmaps, sales, every meeting transcript — into one knowledge base: every brainstorm on a canvas inside the doc, the whole context one MCP call away. Now Claude doesn't show up like an intern you re-brief every morning — it pulls that context and works like an employee who's been here for years.
— Krishnan S G, creator of superlore