Give your AI the context it needs.

The AI Context Protocol (ACP) is an open specification for embedding machine-readable context directly into codebases. It enables AI coding assistants to understand code structure, respect developer constraints, and navigate codebases efficiently — all without requiring dynamic tool connections.

/**
 * @acp:module "Payment Processing"
 * @acp:domain billing
 * @acp:lock restricted — Approval required before modification
 * @acp:owner "payments-team"
 */
function processPayment(amount, currency) {
  // AI tools now understand this is security-critical code
}

The ACP ecosystem consists of several focused repositories:

# Install the CLI
cargo install acp-cli

# Initialize ACP in your project
acp init

# Index your codebase
acp index

# Check constraints before modifying files
acp constraints src/auth/session.ts

🏷️ Structured Annotations — Embed rich metadata directly in code comments using @acp: syntax

🔒 Advisory Constraints — Define lock levels (frozen, restricted, normal) that AI tools respect

📦 Token-Efficient Variables — Reference code elements without verbose inline descriptions

⚡ Pre-Computed Cache — Query codebase structure instantly with indexed JSON

🔌 Universal Integration — Works with Claude, ChatGPT, Copilot, Cursor, and any MCP-compatible tool

┌──────────────────────────────────────────────────────────────────────┐
│                         Your Codebase                                │
│   ┌─────────────────┐  ┌─────────────────┐  ┌─────────────────┐      │
│   │  @acp:module    │  │  @acp:lock      │  │  @acp:domain    │      │
│   │  @acp:owner     │  │  @acp:style     │  │  @acp:depends   │      │
│   └─────────────────┘  └─────────────────┘  └─────────────────┘      │
└──────────────────────────────────────────────────────────────────────┘
                                  │
                                  ▼
                        ┌─────────────────┐
                        │    ACP CLI      │
                        │   acp index     │
                        └─────────────────┘
                                  │
                                  ▼
┌──────────────────────────────────────────────────────────────────────┐
│                     .acp.cache.json                                  │
│   Pre-computed index: symbols, domains, constraints, call graphs     │
└──────────────────────────────────────────────────────────────────────┘
                                  │
           ┌──────────────────────┼──────────────────────┐
           ▼                      ▼                      ▼
   ┌───────────────┐     ┌───────────────┐     ┌───────────────┐
   │ Claude / GPT  │     │    Cursor     │     │   VS Code     │
   │  (via MCP)    │     │ (.cursorrules)│     │  (Extension)  │
   └───────────────┘     └───────────────┘     └───────────────┘

ACP is built on these core principles:

• Advisory, Not Enforced — Constraints guide AI behavior through trust-based collaboration
• Token Efficiency — Minimize context sent to AI systems with pre-computed indexes
• Language Agnostic — Works across any programming language via comment-based annotations
• Incremental Adoption — Add value gradually, no all-or-nothing migration required
• Complementary to MCP — ACP provides static codebase context; MCP handles dynamic external connections

• Discord: Join our server
• Twitter/X: @acp_protocol
• GitHub Discussions: Discussions

We welcome contributions from the community! Whether it's improving documentation, adding language support, or building new integrations — there's a place for you.

• 📖 Read the Specification
• 🐛 Report Issues
• 💡 Propose Features
• 📝 Submit RFCs

All ACP repositories are released under the MIT License.

Give your AI the context it needs.

Specification • CLI • MCP Server • Discord