mindbase — MCP server for Claude Code: an LLM-gardened wiki
Turns papers, articles and notes into a markdown wiki on your disk: the model reads a source, links it to the rest, flags contradictions and writes pages once you approve.
claude mcp add --env MINDBASE_DATA_DIR=<MINDBASE_DATA_DIR> --transport stdio mindbase -- npx -y mindbase-mcpWhat mindbase does
MindBase follows the LLM Wiki idea. Your raw material sits in sources/ and the model never rewrites it. The maintained part — context.md and sources/research/ — changes only after you tick items in the proposed plan. Index, snapshots and lint findings in state/, logs/ and artifacts/ are derived and rebuild from scratch.
Three operations drive it. Ingest walks through a new source and proposes what belongs in the wiki. Build regenerates context.md from everything not yet digested. Lint re-reads the whole wiki and quotes the sentences that contradict each other. That last step is where a vector store falls short: nothing is re-derived at query time, the synthesized answer already lives in the files.
The server starts with npx -y mindbase-mcp, and MINDBASE_DATA_DIR points at your data. The browser app wants Node 20+ and picks a local Ollama model for your RAM, so cloud keys stay optional. In Claude the marketplace plugin adds sub-agents with narrow tool allowlists — the builder gets no file-write tool, only an atomic MCP call.
One catch: in Cursor and Windsurf the model often misses a plain “add to mindbase” request unless you add a routing conventions block to your rules file.
What it helps with
- link scattered articles into one cross-referenced wiki
- surface contradictions between old and new notes
- regenerate
context.mdafter a batch of new sources - keep a personal knowledge base as plain markdown for git and grep
- expose wiki operations to a coding agent over MCP
Example prompts for the agent
- Add this article to mindbase and show me the plan before writing
- Rebuild
context.mdfrom everything not yet ingested - Run a lint pass and quote the claims that conflict
How to install mindbase
Platform CLI
claude mcp add --env MINDBASE_DATA_DIR=<MINDBASE_DATA_DIR> --transport stdio mindbase -- npx -y mindbase-mcpReplace <…> with your own values
Platform CLI
codex mcp add mindbase --env MINDBASE_DATA_DIR=<MINDBASE_DATA_DIR> -- npx -y mindbase-mcpReplace <…> with your own values
Config file
{
"mcpServers": {
"mindbase": {
"command": "npx",
"args": [
"-y",
"mindbase-mcp"
],
"env": {
"MINDBASE_DATA_DIR": "<MINDBASE_DATA_DIR>"
}
}
}
}Replace <…> with your own values
Config file
{
"servers": {
"mindbase": {
"type": "stdio",
"command": "npx",
"args": [
"-y",
"mindbase-mcp"
],
"env": {
"MINDBASE_DATA_DIR": "<MINDBASE_DATA_DIR>"
}
}
}
}Replace <…> with your own values
Platform CLI
gemini mcp add -e MINDBASE_DATA_DIR=<MINDBASE_DATA_DIR> mindbase npx -y mindbase-mcpReplace <…> with your own values
Step by step, with commands for each platform: How to add an MCP server to Claude Code, Codex, Cursor, VS Code, Gemini CLI and more
Questions about mindbase
How do I install mindbase in Claude Code?
From the marketplace: /plugin marketplace add frankchu91/mindbase-llm-wiki first, then /plugin install mb@mindbase. For the bare server use claude mcp add mindbase -- npx -y mindbase-mcp.
What does mindbase do?
Keeps a personal wiki in markdown. It reads a source you add, discusses takeaways, and writes pages plus wikilinks after approval. A separate pass audits the wiki for contradictions and orphans.
What do I need to run it?
Node 20+ and npx -y mindbase-mcp, with MINDBASE_DATA_DIR set to your data path. For the browser app, npx mindbase-app opens a wizard that matches an Ollama model to your RAM.
