MCP 服务器

Sensei

Provides Claude with persistent memory about the user's skills, learning history, and project context to deliver personalized, adaptive explanations and avoid repeating basics.

README

Sensei

AI That Remembers How to Teach You

Claude is smart enough to be a great mentor. It just doesn't know who it's talking to.

Sensei gives Claude memory about you — your skill level, what you've learned, and what you're building.

You: "How do I handle errors in useEffect?"

Claude: [Checks Sensei]
        → Knows you understand promises, are learning React hooks
        → Knows you're building a fintech app with Supabase

Claude: "Since you're comfortable with try/catch from your async work,
        the pattern in useEffect is similar. For your Supabase queries..."

No more "First, let me explain what useEffect is..." every single time.

Quick Start

1. Add to Claude Code

Add this to your Claude Code MCP settings (~/.claude/claude_desktop_config.json):

{
  "mcpServers": {
    "sensei": {
      "command": "npx",
      "args": ["-y", "@sensei-mcp/sensei"]
    }
  }
}

2. Add the System Prompt

Copy the contents of PROMPT.md into your Claude Project's custom instructions.

This tells Claude how to use Sensei effectively.

3. Start Learning

That's it. Claude now remembers who you are.

What Sensei Tracks

Skills

Three levels. Simple.

Level	Meaning
`unknown`	Never discussed
`learning`	Covered basics, still developing
`solid`	Uses independently, understands well

Learning History

What Claude has taught you, so it can reference past explanations:

2025-01-14: explained "useEffect cleanup" (react-hooks)
2025-01-13: explained "async/await" (javascript)

Project Context

What you're building:

Project: fintech-app
Stack: React, TypeScript, Supabase
Notes: "REST API with row-level security"

Prerequisites

Before explaining X, Claude checks if you know the prerequisites:

useEffect requires:
├── react-hooks (concept of hooks)
├── closures (effects close over state)
└── side-effects (what effects are for)

Preferences

How you like explanations:

style: concise | detailed | example-heavy
languages: ["typescript", "python"]

Tools Reference

Tool	Purpose
`sensei_get_context`	Get everything: skills, history, project, preferences
`sensei_get_skills`	Get skill levels
`sensei_set_skill`	Update a skill level
`sensei_log_learned`	Record that a concept was taught
`sensei_get_history`	Get learning history
`sensei_get_prerequisites`	Check prerequisites for a concept
`sensei_get_project`	Get project context
`sensei_set_project`	Update project context
`sensei_get_preferences`	Get explanation preferences
`sensei_set_preferences`	Update preferences

Privacy

All local. Data stored in ~/.sensei/sensei.db
No cloud. No accounts. No telemetry.
Your data. Export or delete anytime.

Development

# Clone
git clone https://github.com/your-username/sensei
cd sensei

# Install
npm install

# Build
npm run build

# Run locally
node dist/index.js

Project Structure

sensei/
├── src/
│   ├── index.ts          # MCP server entry
│   ├── db/               # SQLite database layer
│   │   ├── init.ts       # Database initialization
│   │   ├── skills.ts     # Skills CRUD
│   │   ├── history.ts    # History CRUD
│   │   ├── project.ts    # Project context
│   │   └── preferences.ts
│   └── tools/            # MCP tool handlers
│       ├── index.ts      # Tool registration
│       ├── handlers.ts   # Tool implementations
│       └── prerequisites.ts
├── knowledge/
│   └── prerequisites.json  # Concept prerequisite graph
├── PROMPT.md             # System prompt for Claude
└── README.md

Contributing

The most valuable contributions are to the prerequisite graph.

# Edit knowledge/prerequisites.json
# Add your domain expertise

{
  "kubernetes-pods": ["docker", "containers", "yaml"],
  "graphql-resolvers": ["graphql-schema", "async-await"]
}