v1.0.0 — Self-Building Knowledge Base

🧠 Knowledge Base — Self-Building MCP Server

A personal technical knowledge base that builds itself — collects, distills, and serves via MCP, all from a single container.

✨ Features

• Self-building: daily collect (GitHub Trending + AI news RSS + Anthropic HTML + arXiv) → LLM distill → structured wiki
• Fully local daily loop: collect + distill run inside the container, no GitHub involvement after deploy
• MCP server: search_kb, get_entity, list_recent tools over Streamable HTTP
• Interactive knowledge graph: force-directed visualization at /graph (zoom, pan, search, filter, tag cloud)
• Bearer token auth: optional MCP_AUTH_TOKEN for /mcp endpoint security
• Multi-arch image: linux/amd64 + linux/arm64 (built via GitHub Actions + QEMU)

🔒 Security

• LLM API key lives only in the container environment variable — never touches GitHub
• GitHub hosts source code and builds the image; daily data stays on your machine

📦 Deploy

docker run -d --name knowledge-mcp -p 8000:8000 \
  -e KNOWLEDGE_BASE_PATH=/data/kb-self \
  -e MCP_AUTH_TOKEN=your-secret-token \
  -e LLM_API_BASE=https://api.openai.com/v1 \
  -e LLM_API_KEY=sk-... \
  -e LLM_MODEL=gpt-4o-mini \
  -v /opt:/data:rw \
  ghcr.io/zhengr/knowledge-mcp:latest

Then open http://<host>:8000/graph for the knowledge graph, or connect an MCP client to http://<host>:8000/mcp.

🌐 Languages

• English (default)
• 中文
• 日本語

---

Inspired by mufans/knowledge-base and the Karpathy LLM Wiki method.