ACP: The “New MCP” for IDEs and Coding Agents

Key Concepts:

• Agent Client Protocol (ACP): Standardized communication protocol between code editors/IDEs and coding agents.
• MCP (Mentioned for comparison): Protocol for agent and tool communication.
• Coding Agents: CLI-based tools that assist with coding tasks (e.g., code generation, editing, commenting).
• IDEs (Integrated Development Environments): Software applications that provide comprehensive facilities to computer programmers for software development.
• Diffs: Visual representation of changes between two versions of a file.
• Zed IDE: Open-source IDE developed by Zed Industries, supporting ACP.
• Cloud Code & Gemini CLI: Examples of coding agents that can be integrated with Zed IDE via ACP.
• JSON RPC: Lightweight protocol used for ACP communication.
• Standard IO: Standard input/output streams used for communication.

1. The Problem ACP Solves:

• Currently, integrating coding agents into IDEs is often a "hack" rather than native support.
• Major companies release CLI-based coding agents with limited and inconsistent IDE integration.
• ACP aims to standardize this communication, enabling seamless integration of any coding agent into any supporting IDE.

2. ACP in Action (Zed IDE Example):

• Zed IDE supports ACP and integrates with coding agents like Gemini CLI and Cloud Code.
• Users can interact with coding agents directly within the IDE, without needing separate CLI tools.
• Example: Using Gemini CLI to create a Python file implementing an A* search algorithm.
• Diffs generated by the coding agent are displayed directly within the IDE, allowing for easy review and acceptance of changes.
• Users can switch between different coding agents (e.g., Gemini CLI to Cloud Code) within the same IDE session.

3. Key Benefits of ACP:

• IDE and Agent Agnostic: Any IDE supporting ACP can work with any ACP-compatible coding agent.
• Standardized Communication: Eliminates the need for custom integrations for each IDE-agent combination.
• Reduced Integration Overhead: Simplifies the process of integrating new coding agents into IDEs.
• Universal Integration: Aims to provide a universal way to integrate coding agents, unlike current solutions with limited support.
• First-Class UI: Provides a rich user interface within the IDE, including live diffs, multi-buffer reviews, and navigation.

4. Who is Behind ACP:

• Developed by Zed Industries, the creators of Zed IDE, in conjunction with the Google Gemini team.
• Currently supports Gemini CLI and Cloud Code out of the box.
• Open-source protocol, encouraging community involvement and improvement.

5. Custom Agent Integration:

• ACP allows for the integration of custom agents by creating a JSON file with a specific structure.
• The JSON file specifies the agent server, the command to use, and an index.js file pointing to the agent.

6. Technical Details of ACP:

• Uses a JSON RPC protocol for communication between the code editor and the coding agent.
• Transport agnostic, using standard input/output operations to a subprocess.
• The editor starts the agent and exchanges structured requests and responses.
• The agent proposes actions such as navigation, editing, and running tools.

7. ACP vs. MCP:

• ACP: Controls communication between the editor and the coding agent, enabling control over UI components (e.g., diffs, navigation).
• MCP: Focuses on agent and tool communication, providing a standard way to expose tools or context to the agent.
• ACP and MCP are complementary; an ACP agent can use an MCP server under editor-mandated permissions.
• "Essentially they are complimentary and ACP editor can let an ACP agent use MCP server under editor mandated per permissions."

8. Current Status and Future Considerations:

• ACP is in early stages of development and actively evolving.
• The current implementation is written in Rust (like Zed IDE).
• Adoption by other coding agents and editors is crucial for its success.
• Security and authentication aspects need further development, similar to MCP.

9. Security and Authentication:

• Security and authentication are important considerations for ACP.
• The current implementation is available in Zed IDE, which is an open-source model.

10. Conclusion:

• ACP has the potential to become a significant protocol for standardizing communication between code editors and coding agents.
• It offers benefits such as IDE and agent agnosticism, reduced integration overhead, and a first-class UI experience.
• While still in early development, ACP addresses a key challenge in the evolving landscape of AI-assisted coding.
• The speaker recommends reading about ACP, suggesting it "could be as big as MCP."