Cortex
Obsidian-backed knowledge graph with semantic search, entity extraction, and cross-session memory. 11 MCP tools. Works with Claude Code, Cursor, Windsurf, and any MCP-compatible editor.
README
🧠 Cortex — Proactive Memory Agent for AI Coding Assistants
Obsidian-backed knowledge graph with semantic search, entity extraction, and cross-session memory. 11 MCP tools. Works with Claude Code, Cursor, Windsurf, and any MCP-compatible editor.
What It Does
You: "How does the memory system handle forgetting?"
Cortex: 💡 This reminds me of your previous work on:
1. decisions/memory-architecture-choice — Ebbinghaus decay with 14-day half-life
2. daily/2026-06-19 — You designed the three-tier memory architecture
3. moc/system-architecture — Full system diagram
- Semantic search — Find concepts across languages (EN query → CN content)
- Knowledge graph — 928 entities, 106K relationships, auto-extracted from your notes
- Proactive context —
memory_monitorwatches conversation topics, surfaces relevant memories - Gap analysis — Compare your learning goals against actual knowledge coverage
- Cross-session — Every session loads your memory index + recent daily logs automatically
Architecture
MCP Client (Claude Code / Cursor / Windsurf)
│ stdio JSON-RPC
▼
┌─────────────────────────────────┐
│ Cortex MCP Server (TypeScript) │
│ 11 tools + 1 resource │
│ ┌───────────────────────────┐ │
│ │ SQLite (sql.js) │ │ ← Knowledge graph (8 tables)
│ │ LanceDB │ │ ← Vector embeddings (384-dim)
│ │ ONNX (all-MiniLM-L6-v2) │ │ ← Local, zero API cost
│ └───────────────────────────┘ │
└──────────────┬──────────────────┘
│ reads & indexes
▼
┌─────────────────────────────────┐
│ Obsidian Vault (source of truth)│
│ daily/ decisions/ moc/ │
│ 学习计划/ ... │
└─────────────────────────────────┘
Quick Start
# 1. Install
cd Cortex/mcp-server
npm install --ignore-scripts
npx tsc
# 2. Configure (Claude Code)
# Copy the MCP config to your global Claude Code settings:
# ~/.claude/.mcp.json
# 3. Use
# Start a Claude Code session. The MCP server starts automatically.
# Say: "memory_ingest" to index your vault
# Say: "search for architecture decisions" to find memories
MCP Tools
| Tool | Description |
|---|---|
memory_search |
Hybrid keyword + vector semantic search |
memory_ingest |
Scan vault, extract entities, build index |
memory_status |
System health report |
memory_entity_extract |
Extract typed entities from text |
memory_graph_query |
Query the knowledge graph (multi-hop) |
memory_auto_link |
Suggest [[wikilinks]] between notes |
memory_monitor |
Process conversation, return proactive context |
memory_conflict_resolve |
Detect contradictions with existing memories |
memory_gap_analysis |
Analyze knowledge gaps vs learning goals |
memory_consolidate |
Apply decay, find archive candidates |
memory_session_end |
Generate daily note (portable Stop hook) |
Editor Support
| Editor | Setup |
|---|---|
| Claude Code | ~/.claude/.mcp.json (global) or .mcp.json (project) |
| Cursor | Copy editors/cursor.mcp.json → .cursor/mcp.json |
| Windsurf | Copy editors/windsurf.mcp.json |
| Cline / VS Code | Copy editors/cline.mcp.json |
Project Structure
Cortex/
├── mcp-server/ ← TypeScript MCP server (32 source files)
│ ├── src/
│ │ ├── tools/ ← 11 MCP tools
│ │ ├── graph/ ← Knowledge graph + decay + gap detection
│ │ ├── ingest/ ← Vault scanner, chunker, entity extractor
│ │ ├── embeddings/ ← ONNX pipeline (all-MiniLM-L6-v2)
│ │ ├── db/ ← SQLite + LanceDB
│ │ └── resources/ ← MCP resource endpoints
│ ├── editors/ ← Config templates for other editors
│ └── tests/
├── src/
│ ├── hooks/ ← Bash hooks (SessionStart, Stop, etc.)
│ ├── lib/ ← Bash utilities
│ └── templates/ ← Vault file templates
├── skills/ ← Claude Code skills
├── commands/ ← Slash commands
├── docs/ ← Architecture docs
└── tests/ ← Hook tests
License
MIT — see LICENSE
推荐服务器
Baidu Map
百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。
Playwright MCP Server
一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。
Magic Component Platform (MCP)
一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。
Audiense Insights MCP Server
通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。
VeyraX
一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。
graphlit-mcp-server
模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。
Kagi MCP Server
一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。
e2b-mcp-server
使用 MCP 通过 e2b 运行代码。
Neon MCP Server
用于与 Neon 管理 API 和数据库交互的 MCP 服务器
Exa MCP Server
模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。