VibeTape Server

Record the vibe of your build — A revolutionary Model Context Protocol (MCP) server that captures key development moments, enables multi-agent traceability, provides intelligent context curation, and facilitates seamless agent-to-agent handoffs.

License: MIT TypeScript MCP Version

🚀 What is VibeTape?

VibeTape transforms your development workflow into a proactive context management system. Beyond capturing moments, it provides multi-agent coordination, intelligent context curation, and LangGraph-compatible handoffs that work with any MCP-compatible AI client.

Perfect for:

• 🤖 Multi-agent systems that need shared context and traceability
• 🎯 Solo developers who want to remember past solutions
• 👥 Teams who need shared knowledge and context
• 🔄 AI orchestration frameworks (LangGraph, CrewAI, AutoGen)
• 📚 Technical leads building institutional knowledge

✨ Key Features

🤖 Multi-Agent Traceability (NEW v0.4.0)

• Actor management — Register and track humans and AI agents
• Task lifecycle — Create, assign, and hand off tasks between agents
• Agent analytics — Success rates, activity patterns, performance metrics
• Temporal tracking — Know when facts became true and when they were superseded

🧠 Intelligent Context Curation (NEW v0.4.0)

• RankRAG-style scoring — Relevance scoring with weighted factors
• Context window optimization — Fit the best context within token budgets
• Agent needs prediction — Anticipate what context an agent will need
• Smart moment selection — Balance relevance, recency, and signal quality

🔄 Agent-to-Agent Handoffs (NEW v0.4.0)

• LangGraph-compatible payloads — Direct integration with agent frameworks
• RETEX-aware handoffs — Include relevant lessons learned
• Risk warnings — Highlight potential issues for receiving agents
• Task continuity — Seamless work transfer between agents

🚀 Context Handoff System (v0.3.0)

• Transition cards — Generate compact context summaries (350 tokens)
• Smart ranking — Intelligent moment prioritization by recency, type, and impact
• Cross-session continuity — Never lose context between AI sessions
• Proactive suggestions — Auto-detect when context window is saturating

🧹 Intelligent Denoising (v0.3.0)

• Noise filtering — Auto-detect and filter trivial moments
• Duplicate merging — Consolidate similar entries intelligently
• Signal scoring — Quality metrics for moment relevance (0-1 scale)

🎯 Smart Moment Capture

• Wins, fails, decisions, notes — capture what matters
• Git context — automatic branch, commit, and diff tracking
• Actor attribution — Know who (human or AI) created each moment

🔍 Intelligent Search

• Semantic search with OpenAI embeddings (TF-IDF fallback)
• Advanced filtering by tags, dates, types, and regex
• Relation mapping — link related moments (causes, solves, relates, supersedes)

🧠 AI-Powered Insights

• RETEX cards — AI-generated prescriptive rules from your experiences
• Task-aware RETEX — Get relevant lessons for specific tasks
• Pattern detection — find recurring issues automatically

🏃‍♂️ Quick Start

1. Install

git clone https://github.com/sambaleuk/Vibetape-MCP-Server.git
cd Vibetape-MCP-Server
npm install
npm run build

2. Configure Your AI Client

VibeTape works with any MCP-compatible AI client:

🤖 Claude Desktop

Add to your claude_desktop_config.json:

{
  "mcpServers": {
    "vibetape": {
      "command": "node",
      "args": ["/absolute/path/to/Vibetape-MCP-Server/dist/server.js"],
      "cwd": "/absolute/path/to/Vibetape-MCP-Server",
      "env": {
        "OPENAI_API_KEY": "your-openai-key-here"
      }
    }
  }
}

💻 Cursor IDE

Add to your ~/.cursor/mcp.json:

{
  "vibetape": {
    "command": "node",
    "args": ["--loader", "ts-node/esm", "src/server.ts"],
    "cwd": "/absolute/path/to/Vibetape-MCP-Server",
    "env": {
      "OPENAI_API_KEY": "your-openai-key-here"
    }
  }
}

🔧 Continue.dev / Other MCP Clients

VibeTape implements the full MCP specification and works with any compliant client.

3. Start Using

Restart your AI client and start capturing moments:

Hey AI, mark this moment: "Successfully implemented Redis caching" as a win with tags: api, performance

4. Multi-Agent Example (v0.4.0)

# Register an agent
register_actor with id: "code_reviewer", type: "agent", name: "Code Reviewer"

# Create a task
create_task with title: "Review authentication module", assigned_to: "code_reviewer"

# Agent captures moments linked to the task
mark_moment with title: "Found SQL injection vulnerability", task_id: "task_xyz"

# Hand off to another agent
create_handoff_for_agent with task_id: "task_xyz", from_agent: "code_reviewer", to_agent: "security_fixer"

🛠️ Available Tools

🤖 Multi-Agent Tools (NEW v0.4.0)

• register_actor — Register a human or AI agent
• get_actor — Get actor details and capabilities
• list_actors — List all registered actors
• get_actor_stats — Get performance statistics for an actor
• create_task — Create a new task with assignment
• update_task — Update task status and outcome
• list_tasks — List tasks with filtering options
• get_task_context — Get all moments related to a task

🧠 Context Intelligence Tools (NEW v0.4.0)

• context_relevance_score — Calculate RankRAG-style relevance for moments
• evaluate_context_window — Optimize context selection within token budget
• predict_agent_needs — Predict what context an agent will need
• get_retex_for_task — Get relevant RETEX cards for a task
• create_handoff_for_agent — Create LangGraph-compatible handoff payload

🚀 Context Handoff Tools (v0.3.0)

• generate_context_handoff — Create compact transition cards (350 tokens)
• suggest_transition_card — Auto-suggest handoff when context saturates
• sweep_noise — Intelligent denoising of trivial/duplicate moments

Core Tools

• mark_moment — Capture key development moments (now with actor_id, task_id)
• search_moments — Find similar past experiences
• list_moments — Browse recent captures
• make_retex — Generate AI prescriptive cards
• export_timeline — Day-by-day development timeline

Advanced Tools

• link_moments — Create relationships between moments
• supersede_moment — Mark a moment as superseded by another (temporal tracking)
• comment_moment — Add collaborative annotations
• search_moments_advanced — Multi-criteria search
• stats_overview — Development pattern analytics

📋 Resources

🤖 Agent Resources (NEW v0.4.0)

• actor://{id} — Actor details with stats (JSON)
• task://{id} — Task details with related moments (JSON)

🚀 Context Handoff Resources

• handoff://{id} — Transition card for cross-session continuity (Markdown)

Core Resources

• moment://{id} — Individual moment details (JSON)
• timeline://{day} — Daily timeline (Markdown)
• retex://{id} — AI-generated prescriptive card (JSON)
• graph://{id} — Moment relationship graph (JSON)

🔧 Configuration

Environment Variables

# Optional: OpenAI for semantic search and RETEX generation
OPENAI_API_KEY=sk-your-key-here

# Optional: Custom storage location (default: ~/.vibetape)
VIBETAPE_HOME=~/.vibetape

# Optional: Team collaboration directory
VIBETAPE_TEAM_DIR=~/your-team-repo

Works Without OpenAI

VibeTape gracefully degrades without OpenAI:

• ✅ TF-IDF semantic search (good for most cases)
• ✅ All multi-agent features work fully
• ❌ No AI-generated RETEX cards

🏗️ Architecture

VibeTape follows MCP (Model Context Protocol) standards and is designed for multi-agent orchestration:

┌─────────────────┐    ┌─────────────────┐    ┌─────────────────┐
│  MCP Clients    │◄──►│  VibeTape MCP   │◄──►│  Local Storage  │
│ Claude/Cursor/  │    │     Server      │    │   ~/.vibetape   │
│ LangGraph/CrewAI│    │    (v0.4.0)     │    │ + Team Vault    │
└─────────────────┘    └─────────────────┘    └─────────────────┘
         │                     │
         │                     ▼
         │            ┌─────────────────┐
         │            │   OpenAI API    │
         │            │   (optional)    │
         │            └─────────────────┘
         │
         ▼
┌─────────────────────────────────────────┐
│         Multi-Agent Orchestration        │
│  ┌─────────┐  ┌─────────┐  ┌─────────┐  │
│  │ Agent A │─►│ Handoff │─►│ Agent B │  │
│  │(reviewer)│  │ Payload │  │ (fixer) │  │
│  └─────────┘  └─────────┘  └─────────┘  │
└─────────────────────────────────────────┘

Agent Handoff Flow (v0.4.0)

Agent A (Code Reviewer)     VibeTape           Agent B (Security Fixer)
     │                          │                       │
     ├─► create_task ──────────►│                       │
     ├─► mark_moment (findings)─►│                       │
     │                          │                       │
     ├─► create_handoff_for_agent─►│                    │
     │   (LangGraph payload)    │                       │
     │                          │                       │
     │                          │   ◄── read handoff ──┤
     │                          │                       │
     │                          └─► Full context ──────►│
     │                              + RETEX cards       │
     │                              + Risk warnings     │

📊 Use Cases

Multi-Agent Development Pipeline

# Code reviewer agent finds issues
register_actor id: "reviewer", type: "agent"
create_task title: "Security audit of auth module"
mark_moment title: "Found 3 SQL injection vulnerabilities"

# Hand off to security agent
create_handoff_for_agent from: "reviewer", to: "security_fixer"
→ Includes context, RETEX cards, risk warnings

# Security agent fixes and reports
update_task status: "completed", outcome: "success"

Context-Aware Agent Routing

# Predict what context an agent needs
predict_agent_needs task_id: "xxx", actor_id: "debugger"
→ Returns recommended moments, RETEX cards, warnings

# Evaluate optimal context window
evaluate_context_window task_id: "xxx", budget_tokens: 2000
→ Returns ranked moments that fit the budget

Cross-Session Continuity

# End of day in Claude Desktop
Generate handoff → Get compact transition card

# Next morning in Cursor IDE
Read handoff://{id} → Instantly resume with full context

🔒 Security & Privacy

• 🔐 Local storage only — Data stays in ~/.vibetape/ by default
• 👀 Read-only project access — Never modifies your code
• 🚫 No shell execution — Only safe Git read operations
• 🌐 Minimal network — Only OpenAI API (optional)
• 🔑 Environment variables — API keys never hardcoded

📈 Roadmap

🚀 Future Features

• SQLite backend — Better performance for large datasets
• Web dashboard — Visual relationship graphs and analytics
• Native LangGraph integration — Direct Command pattern support
• VS Code extension — Native IDE integration
• Export integrations — Notion, Obsidian, etc.

🤝 Contributing

We welcome contributions! See CONTRIBUTING.md for guidelines.

1. Fork the repository
2. Create a feature branch (git checkout -b feature/amazing-feature)
3. Commit your changes (git commit -m 'Add amazing feature')
4. Push to the branch (git push origin feature/amazing-feature)
5. Open a Pull Request

📝 License

This project is licensed under the MIT License - see the LICENSE file for details.

🙏 Acknowledgments

• Built with Model Context Protocol (MCP) by Anthropic
• Inspired by multi-agent orchestration frameworks (LangGraph, CrewAI)
• Thanks to the open source community for amazing tools and libraries

---

Ready to orchestrate your AI agents? ⭐ Star this repo and start building!

Get Started • Join Discussions • Report Issues