Build a Persistent LLM Wiki Using Claude and the Model Context Protocol

How to build an LLM wiki with How to build an LLM wiki with Claude and MCP

Developer Andrej Karpathy proposed the LLM wiki pattern, but local setup complexity often prevents adoption. By using Hjarni’s hosted MCP server, engineers can connect Claude to a persistent knowledge base in just ten minutes without local terminal configuration.

Why This Matters

Traditional LLM interactions rely on one-shot context where users must repeatedly paste relevant documentation or preferences, leading to context reset and inefficiency. This MCP-driven approach allows for a technical reality where the AI reads from and writes back to a structured repository, ensuring architectural decisions and research summaries remain accessible across different sessions and platforms. This eliminates the friction of manual bookkeeping and ensures that context compounds over time rather than resetting with every new chat window.

Key Insights

• Persistent context via MCP allows Claude to fetch notes on-demand rather than relying on static token window pastes.
• The Write-back loop enables Claude to update notes, such as project decisions or research summaries, directly into the Hjarni knowledge base.
• Hjarni serves as a hosted MCP server, removing the requirement for local Obsidian vaults or terminal-based Claude Code setups.
• Cross-client compatibility ensures both Claude Desktop and ChatGPT can access and modify the same repository through MCP.
• Automated structure setup using the Knowledge Wiki template provides AI instructions that define how Claude should maintain the wiki.

Working Examples

Recommended folder structure for a starter LLM wiki.

/About me
- Profile
- Preferences
- Current goals
/Stack
- Languages and frameworks
- Tools
- Architecture decisions
/Projects
/Project A
- Overview
- Decisions
- Open questions
/Research
- Articles and summaries
- Things to revisit

Example of a persistent ‘About me’ note used for cross-session context.

# About me
I'm a solo developer building an AI-native knowledge base.
## Current projects
- Product development
- Growth and content
- Customer feedback
## Preferences
- Prefer short, direct answers
- Ask before making large architectural changes
- Save useful context back to my wiki
## Current goals
- Publish more useful content
- Improve onboarding
- Make my AI context reusable across sessions

Practical Applications

• Architectural Bookkeeping: Use Claude to automatically update ‘Stack’ notes when switching tools to maintain live documentation. Pitfall: Manual maintenance leads to stale docs; letting the AI handle updates ensures the wiki remains current.
• Research Compounding: Synthesize multiple sources in one session and save to a ‘Research’ folder for Claude to reference in future sessions. Pitfall: One-shot pasting results in lost research context once the chat session ends.

References:

• https://dev.to/hjarni/how-to-build-an-llm-wiki-with-how-to-build-an-llm-wiki-with-claude-and-mcp-52m4