Client implementation for the MCP SDK.

This crate provides the client-side implementation for the Model Context Protocol. It includes a fluent client API, server discovery, and connection management.

Overview

The MCP client allows AI applications to:

• Connect to MCP servers via various transports
• Discover and invoke tools
• Read resources
• Get prompts
• Track long-running tasks

Example

use mcpkit_client::{Client, ClientBuilder};
use mcpkit_transport::SpawnedTransport;

#[tokio::main]
async fn main() -> Result<(), mcpkit_core::error::McpError> {
    // Spawn an MCP server as a subprocess and connect via stdio
    let transport = SpawnedTransport::spawn("my-mcp-server", &[] as &[&str]).await?;

    let client = ClientBuilder::new()
        .name("my-client")
        .version("1.0.0")
        .build(transport)
        .await?;

    // List available tools
    let tools = client.list_tools().await?;
    for tool in &tools {
        println!("Tool: {}", tool.name);
    }

    // Call a tool
    let result = client.call_tool("add", serde_json::json!({
        "a": 1,
        "b": 2
    })).await?;

    Ok(())
}

Client Handler

For handling server-initiated requests (sampling, elicitation), implement the [ClientHandler] trait:

use mcpkit_client::ClientHandler;
use mcpkit_core::types::{CreateMessageRequest, CreateMessageResult};
use mcpkit_core::error::McpError;

struct MyHandler;

impl ClientHandler for MyHandler {
    // Override default methods as needed
}