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
🧠 Smart Kernel v3.1	High-speed boot with auto-fix for stale context, broken locks, and missing data
🔌 Multi-Agent Support	Works with 8+ AI assistants via unified adapter
🤝 Session Handoffs	Incomplete work is saved—next session picks it up
🔄 Auto-Maintenance	Agents self-correct: break stale locks, trigger audits, prune old learnings
📂 Monorepo Ready	Define sub-projects with scoped contexts
🧠 Knowledge Persistence	Patterns, gotchas, and decisions survive across sessions
⚡ Context Budget System	Tiered loading prevents LLM context overload
🔁 `/reset` Command	Clear working context and start fresh

🔧 Supported AI Assistants

See Universal Adapter.

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, development, maintenance...
│   └── adapters/     # Per-agent integration guides
└── archive/          # Historical context storage

🚀 Quick Start

1. Install

Via CLI (Recommended):

npx corepackai install @corepackai/ucp
# OR for a full team setup:
npx corepackai install @corepackai/startup-team

Or Manual: Copy the pack into your project's context directory (e.g., .context/ or .ai/).

2. Initialize

Tell your agent:

"Read bin/workflows/onboarding.md"

3. Audit

Tell your agent:

"Read bin/workflows/audit.md and audit this project"

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

📖 Documentation

Resource	Description
User Guide	Detailed setup and usage for humans
boot.md	Agent entry point (boot protocol, structure)
CLI Guide	Automation tool instructions

Installation

To install this package, run:

npx corepackai install @corepackai/ucp --bridge

Unified Context Protocol