Struct McpClient 

pub struct McpClient<T: McpTransport> { /* private fields */ }

Available on crate feature mcp only.

MCP client for communicating with MCP servers.

The client handles the MCP protocol, including initialization, tool discovery, and tool execution.

Example

use agent_sdk::mcp::{McpClient, StdioTransport};

// Spawn server and create client
let transport = StdioTransport::spawn("npx", &["-y", "mcp-server"]).await?;
let client = McpClient::new(transport, "my-server".to_string()).await?;

// List available tools
let tools = client.list_tools().await?;

// Call a tool
let result = client.call_tool("tool_name", json!({"arg": "value"})).await?;

Implementations

impl<T: McpTransport> McpClient<T>

pub async fn new(transport: Arc<T>, server_name: String) -> Result<Self>

Create a new MCP client and initialize the connection.

Arguments

• transport - The transport to use for communication
• server_name - A name to identify this server connection

Errors

Returns an error if initialization fails.

pub const fn new_uninitialized(transport: Arc<T>, server_name: String) -> Self

Create a client without initialization.

Use this if you need to control when initialization happens.

pub async fn initialize(&mut self) -> Result<&InitializeResult>

Initialize the MCP connection.

This must be called before using other methods.

Errors

Returns an error if the server rejects initialization.

pub fn server_name(&self) -> &str

Get the server name.

pub const fn server_info(&self) -> Option<&InitializeResult>

Get server info if initialized.

pub fn protocol_version(&self) -> Option<&str>

The MCP protocol revision negotiated with the server.

Returns None until McpClient::initialize has completed. This is the revision the server selected, which may be older than PREFERRED_PROTOCOL_VERSION if the server is on a legacy build.

pub async fn list_tools(&self) -> Result<Vec<McpToolDefinition>>

List available tools from the server.

Errors

Returns an error if the request fails.

pub async fn call_tool( &self, name: &str, arguments: Value, ) -> Result<McpToolCallResult>

Call a tool on the server.

Arguments

• name - Tool name to call
• arguments - Tool arguments as JSON

Errors

Returns an error if the tool call fails.

pub async fn call_tool_raw( &self, name: &str, arguments: Option<Value>, ) -> Result<McpToolCallResult>

Call a tool with raw Value arguments.

Arguments

• name - Tool name to call
• arguments - Tool arguments as optional JSON

Errors

Returns an error if the tool call fails.

pub async fn list_resources(&self) -> Result<Vec<McpResource>>

List resources exposed by the server (resources/list).

Resources are addressable data (files, database rows, API payloads) the server makes available for reading. Returns an empty list if the server did not advertise the resources capability.

Errors

Returns an error if the request fails or the response cannot be parsed.

pub async fn read_resource(&self, uri: &str) -> Result<ResourceReadResult>

Read a resource by URI (resources/read).

Errors

Returns an error if the request fails or the response cannot be parsed.

pub async fn list_prompts(&self) -> Result<Vec<McpPrompt>>

List prompts exposed by the server (prompts/list).

Returns an empty list if the server did not advertise the prompts capability.

Errors

Returns an error if the request fails or the response cannot be parsed.

pub async fn get_prompt( &self, name: &str, arguments: Option<Value>, ) -> Result<PromptGetResult>

Fetch and render a prompt by name (prompts/get).

Arguments

• name - Prompt name to fetch.
• arguments - Optional arguments to interpolate into the template.

Errors

Returns an error if the request fails or the response cannot be parsed.

pub fn supports_resources(&self) -> bool

Whether the server advertised the resources capability.

pub fn supports_prompts(&self) -> bool

Whether the server advertised the prompts capability.

pub async fn close(&self) -> Result<()>

Close the client connection.

Errors

Returns an error if the transport fails to close.