In April 2026, Andrej Karpathy published his LLM wiki: markdown files an agent maintains, sitting between you and your raw sources.
In June, Google published the Open Knowledge Format for the same idea: one page per concept, plain markdown with frontmatter, readable by a person or an agent without special tooling.
Technical teams used these LLM-wikis for playbooks, onboarding, process documentation, and more. Pages stay up-to-date, you don't have to curate Google Docs, Notion, or Confluence, and your AI gets smarter as the wiki grows.
But nobody outside the technical team could use it.
The problem: these wikis lives in a GitHub repository. Using it means knowing Claude Code or Codex, Git, and pull requests.
Isomorphic removes all that. Set up the connector and your AI talks to the wiki directly via Model Context Protocol (MCP). Nobody clones a repo, nobody opens GitHub, and there is no second app to log into. It's all right there in your chat window.
Your team owns all the data and knowledge: one markdown file per page, in full version history, readable in your GitHub whenever you want.
Isomorphic is just the bridge. Simple as.
Hey Claude, show me our knowledge base.
Paste the endpoint into any assistant that speaks Model Context Protocol. In Claude that is Settings → Connectors → Add custom connector. Sign in by email, then ask it to create your first wiki.
https://mcp.isomorphic.sh/mcp