Personal RAG Knowledge Base

A production-ready personal RAG system over 5,000+ knowledge documents — semantic search across notes, meetings, conversations, and tickets, accessible from Claude Desktop, Claude.ai web, and iOS app via the Model Context Protocol.

M1 done P0 Size S Foundation

📅 Apr 27, 2026 · Cập nhật Apr 29, 2026

5,329 Sources notes · meetings · tickets

47K+ Chunks VECTOR(384) HNSW

1.16s p95 latency embed → ANN → tunnel

0.85+ Top-hit score multilingual e5

~22h Build time over 2 days

Python 3.10 MCP SDK FastMCP Starlette

sentence-transformers

multilingual-e5-small

Oracle ADB 23ai VECTOR(384) HNSW

Cloudflare Tunnel OAuth 2.0 (PKCE + DCR)

OCI Object Storage

A production-ready personal RAG system. Semantic search across thousands of personal knowledge documents directly from Claude Desktop, Claude.ai web, or Claude iOS app — via the Model Context Protocol.

At a glance

5,000+ sources / 47,000+ chunks indexed (mixed Vietnamese + English content)
Multilingual semantic search — top-hit similarity score 0.85+ on real queries
End-to-end p95 latency: 1.16s (embed query → ANN search → tunnel)
MCP Streamable HTTP server with OAuth 2.0 (PKCE + DCR) + legacy bearer fallback
Persistent HTTPS endpoint via Cloudflare named tunnel
Auto-ingest pipeline — files written to a local KB folder are automatically chunked, embedded, and indexed via idempotent SHA-256 hash check
Weekly backup to OCI Object Storage via write-only Pre-authenticated Request (zero credentials on the serving VM)
Filesystem-first — markdown files are the canonical source of truth; the vector DB is a rebuildable derived index

Stack

Python 3.10 · MCP SDK 1.27 · FastMCP · Starlette + uvicorn · sentence-transformers · multilingual-e5-small · Oracle ADB 23ai (VECTOR type + HNSW) · Cloudflare Tunnel · OCI Object Storage · systemd · launchd

Documentation

Doc	Read this for
PRD	What & why — problem framing, goals, scope, milestones, success metrics
Architecture	System diagrams, data flows, component responsibilities, failure modes
Implementation	Tech stack, code structure, schema, performance numbers, security model, reproducibility steps
Notes	Chronological decision log + gotchas + working-session hours

Quickstart for clients

Claude Desktop (macOS):

// ~/Library/Application Support/Claude/claude_desktop_config.json
{
  "mcpServers": {
    "personal-rag-kb": {
      "command": "npx",
      "args": ["-y", "mcp-remote", "https://<your-host>/mcp",
               "--header", "Authorization:${AUTH_HEADER}"],
      "env": { "AUTH_HEADER": "Bearer <your-token>" }
    }
  }
}

Claude.ai web / iOS app: Settings → Connectors → Add Custom Connector → enter your MCP URL → complete OAuth login.

Project status

Day	Milestone
1	MCP server scaffold + Cloudflare tunnel + bearer auth
2	`kb_health` / `kb_ingest` / `kb_search` / `kb_stats` tools + ADB schema
3	Bulk migrate 5,000+ sources / 45K chunks
3b	Multilingual upgrade: BGE-en → multilingual-e5-small
4-5	Sync workflow refactor — 4 source types auto-ingest
6	Persistent named tunnel on a custom domain
7	Weekly backup + disaster recovery script
8	OAuth 2.0 — Claude.ai web + iOS app access

Total build time: ~22 hours over 2 days.

Foundation for downstream projects

This RAG infrastructure is designed as a shared foundation for multiple downstream AI side-projects: recipe extractor, research agent, support bot, meeting summarizer, finance advisor, health coach. Each can call kb_ingest / kb_search rather than building its own vector pipeline.