Unified Context Protocol
# Unified Context Protocol (UCP)
**A drop-in context system that makes AI coding assistants *actually remember* your project.**
## 🎯 The Problem
AI coding assistants forget everything between sessions. You waste time re-explaining your project architecture, coding standards, and past decisions—and agents often create conflicting or inconsistent work.
## 💡 The Solution
UCP is a **standardized governance layer** that sits between your codebase and AI agents. It provides a file-system-based protocol for persistent context, enabling agents to:
- **Read** project state and history
- **Write** learnings and decisions
- **Coordinate** across sessions and tools
## ✨ Key Features
| Feature | Description |
|---------|-------------|
| **🚀 Boot Protocol** | Agents run a startup checklist to load context automatically |
| **🔌 Multi-Agent Support** | Works with 8+ AI assistants (Cursor, Claude, Copilot, Windsurf, Aider, and more) |
| **🤝 Session Handoffs** | Incomplete work is saved—next session picks it up, even with a different agent |
| **📂 Monorepo Ready** | Define sub-projects with scoped contexts |
| **🧠 Knowledge Persistence** | Patterns, gotchas, and decisions survive across sessions |
| **📋 Pre-Built Workflows** | High-performance pseudo-code for audits, features, bugfixes, refactoring, and maintenance |
| **⚡ Context Budget System** | Tiered loading prevents LLM context overload |
| **🎓 Agent Learning** | Tracks user preferences, what worked/failed, and external dependencies |
## 🔧 Supported AI Assistants
Cursor • Windsurf • Claude Code • GitHub Copilot • JetBrains AI • Aider • Gemini Antigravity • Generic LLMs
## 📂 What's Included
```
unified_context_protocol/
├── boot.md # Agent entry point
├── context/ # Project state (MASTER.md, tech.md, changelog.md...)
├── knowledge/ # Persistent learnings & decisions
├── bin/
│ ├── workflows/ # audit, feature, bugfix, refactor, maintenance...
│ └── adapters/ # Per-agent integration guides
└── archive/ # Historical context storage
```
## 🚀 Quick Start
1. **Install** the pack into your project's context directory
2. **Onboard** your agent: `"Read bin/workflows/onboarding.md"`
3. **Audit** your project: `"Read bin/workflows/audit.md and audit this project"`
4. **Done!** Your agents now have persistent, coordinated context
## 🏆 Why UCP?
- **Escape vendor lock-in**: Your context lives in version-controlled files, not proprietary cloud systems
- **Tool-agnostic**: Switch between AI assistants without losing context
- **Human-readable**: All files are markdown—you can edit them directly
- **Version-controlled**: Commit your context alongside your code
