Crate agent_client_protocol

Expand description

§Agent Client Protocol (ACP)

The Agent Client Protocol standardizes communication between code editors (IDEs, text-editors, etc.) and coding agents (programs that use generative AI to autonomously modify code).

§Protocol & Transport

ACP is a JSON-RPC based protocol. While clients typically start agents as subprocesses and communicate over stdio (stdin/stdout), this crate is transport-agnostic.

You can use any bidirectional stream that implements AsyncRead and AsyncWrite.

§Core Components

Agent: Programs that use generative AI to autonomously modify code
- See: Agent
Client: Code editors that provide the interface between users and agents
- See: Client

§Getting Started

To understand the protocol, start by exploring the Agent and Client traits, which define the core methods and capabilities of each side of the connection.

To see working examples of these traits in action, check out the agent and client example binaries included with this crate.

§Implementation Pattern

ACP uses a symmetric design where each participant implements one trait and creates a connection that provides the complementary trait:

Agent builders implement the Agent trait to handle client requests (like initialization, authentication, and prompts). They pass this implementation to AgentSideConnection::new, which returns a connection providing Client methods for requesting permissions and accessing the file system.
Client builders implement the Client trait to handle agent requests (like file system operations and permission checks). They pass this implementation to ClientSideConnection::new, which returns a connection providing Agent methods for managing sessions and sending prompts.

For the complete protocol specification and documentation, visit: https://agentclientprotocol.com

Structs§

AgentCapabilities: Capabilities supported by the agent.
AgentMethodNames: Names of all methods that agents handle.
AgentSide: Marker type representing the agent side of an ACP connection.
AgentSideConnection: An agent-side connection to a client.
Annotations: Optional annotations for the client. The client can use annotations to inform how objects are used or displayed
AudioContent: Audio provided to or from an LLM.
AuthMethod: Describes an available authentication method.
AuthMethodId: Unique identifier for an authentication method.
AuthenticateRequest: Request parameters for the authenticate method.
BlobResourceContents: Binary resource contents.
CancelNotification: Notification to cancel ongoing operations for a session.
ClientCapabilities: Capabilities supported by the client.
ClientMethodNames: Names of all methods that clients handle.
ClientSide: Marker type representing the client side of an ACP connection.
ClientSideConnection: A client-side connection to an agent.
Diff: A diff representing file modifications.
EmbeddedResource: The contents of a resource, embedded into a prompt or tool call result.
EnvVariable: An environment variable to set when launching an MCP server.
Error: JSON-RPC error object.
ErrorCode: Predefined error codes for common JSON-RPC and ACP-specific errors.
FileSystemCapability: File system capabilities that a client may support.
ImageContent: An image provided to or from an LLM.
InitializeRequest: Request parameters for the initialize method.
InitializeResponse: Response from the initialize method.
LoadSessionRequest: Request parameters for loading an existing session.
McpServer: Configuration for connecting to an MCP (Model Context Protocol) server.
NewSessionRequest: Request parameters for creating a new session.
NewSessionResponse: Response from creating a new session.
PermissionOption: An option presented to the user when requesting permission.
PermissionOptionId: Unique identifier for a permission option.
Plan: An execution plan for accomplishing complex tasks.
PlanEntry: A single entry in the execution plan.
PromptCapabilities: Prompt capabilities supported by the agent in session/prompt requests.
PromptRequest: Request parameters for sending a user prompt to the agent.
PromptResponse: Response from processing a user prompt.
ProtocolVersion: Protocol version identifier.
ReadTextFileRequest: Request to read content from a text file.
ReadTextFileResponse: Response containing the contents of a text file.
RequestPermissionRequest: Request for user permission to execute a tool call.
RequestPermissionResponse: Response to a permission request.
ResourceLink: A resource that the server is capable of reading, included in a prompt or tool call result.
SessionId: A unique identifier for a conversation session between a client and agent.
SessionNotification: Notification containing a session update from the agent.
StreamMessage: A message that flows through the RPC stream.
StreamReceiver: A receiver for observing the message stream.
TextContent: Text provided to or from an LLM.
TextResourceContents: Text-based resource contents.
ToolCall: Represents a tool call that the language model has requested.
ToolCallId: Unique identifier for a tool call within a session.
ToolCallLocation: A file location being accessed or modified by a tool.
ToolCallUpdate: An update to an existing tool call.
ToolCallUpdateFields: Optional fields that can be updated in a tool call.
WriteTextFileRequest: Request to write content to a text file.

Enums§

AgentNotification: All possible notifications that an agent can send to a client.
AgentRequest: All possible requests that an agent can send to a client.
AgentResponse: All possible responses that an agent can send to a client.
ClientNotification: All possible notifications that a client can send to an agent.
ClientRequest: All possible requests that a client can send to an agent.
ClientResponse: All possible responses that a client can send to an agent.
ContentBlock: Content blocks represent displayable information in the Agent Client Protocol.
EmbeddedResourceResource: Resource content that can be embedded in a message.
PermissionOptionKind: The type of permission option being presented to the user.
PlanEntryPriority: Priority levels for plan entries.
PlanEntryStatus: Status of a plan entry in the execution flow.
RequestPermissionOutcome: The outcome of a permission request.
Role: The sender or recipient of messages and data in a conversation.
SessionUpdate: Different types of updates that can be sent during session processing.
StopReason: Reasons why an agent stops processing a prompt turn.
StreamMessageContent: The content of a stream message.
StreamMessageDirection: The direction of a message in the RPC stream.
ToolCallContent: Content produced by a tool call.
ToolCallStatus: Execution status of a tool call.
ToolKind: Categories of tools that can be invoked.

Constants§

AGENT_METHOD_NAMES: Constant containing all agent method names.
CLIENT_METHOD_NAMES: Constant containing all client method names.
V0
V1
VERSION

Traits§

Agent: Defines the interface that all ACP-compliant agents must implement.
Client: Defines the interface that ACP-compliant clients must implement.