Llm Wiki Mcp
Mount a Karpathy-style growing markdown wiki so your agent reads pages and appends structured notes to a persistent log over sessions.
Overview
LLM Wiki MCP is a MCP server for the Build phase that exposes a Karpathy-style markdown wiki on disk so agents read pages and append to a persistent log.
What is this MCP server?
- stdio MCP server via PyPI package llm-wiki-mcp (v0.1.1)
- Required --wiki-root filepath: pages under pages/ and append-only log.md
- Karpathy-style LLM wiki pattern: markdown pages agents extend over time
- uvx runtime hint for quick Python invocation
- Persistent project knowledge outside a single chat window
- Server version 0.1.1
- PyPI package llm-wiki-mcp with uvx runtime hint
- Required named argument --wiki-root (filepath)
What problem does it solve?
Important research and decisions evaporate when chat sessions end because there is no structured, agent-addressable knowledge store on your machine.
Who is it for?
Builders running long-horizon agent projects who want a local markdown wiki the model can maintain through MCP.
Skip if: Teams needing multi-user auth, rich WYSIWYG editing, or hosted knowledge bases without local filesystem setup.
What do I get? / Deliverables
After configuration with --wiki-root, your agent can query wiki pages and append to log.md so project knowledge compounds across sessions.
- Agent-readable wiki pages under <wiki-root>/pages/
- Append-only session log at <wiki-root>/log.md
- Repeatable MCP access to growing project knowledge
Recommended MCP Servers
Journey fit
Persistent agent memory belongs in Build when you are shaping how the product and the agent share context across features. Agent-tooling is the right shelf because the server wires filesystem wiki layout into MCP tools rather than shipping end-user UI.
How it compares
Filesystem-backed wiki MCP, not a hosted Notion integration or a RAG-only vector store skill.
Common Questions / FAQ
Who is LLM Wiki MCP for?
Solo developers and agent-heavy workflows that want persistent, markdown-first project wikis the assistant can update via MCP.
When should I use LLM Wiki MCP?
Use it during Build when you are iterating on a product and need durable notes, research, and changelogs the same agent can read and extend.
How do I add LLM Wiki MCP to my agent?
Install llm-wiki-mcp from PyPI (or uvx), register the MCP server with stdio transport, and pass the required --wiki-root absolute path to your wiki directory containing pages/ and log.md.