Claude Code SDK for Rust

Rust SDK for Claude Code. See the Claude Code SDK documentation for more information.

Installation

Add this to your Cargo.toml:

[dependencies]
claude-code-sdk = "0.0.3"
tokio = { version = "1.0", features = ["full"] }

Prerequisites:

• Rust 1.70+
• Node.js
• Claude Code: npm install -g @anthropic-ai/claude-code

Quick Start

use claude_code_sdk::query;
use tokio_stream::StreamExt;

#[tokio::main]
async fn main() -> Result<(), Box<dyn std::error::Error>> {
    // Optional: Initialize logging to see detailed SDK operations
    claude_code_sdk::init_tracing();
    
    let mut stream = query("What is 2 + 2?", None).await?;
    
    while let Some(message) = stream.next().await {
        println!("{:?}", message);
    }
    
    Ok(())
}

Usage

Basic Query

use claude_code_sdk::{query, ClaudeCodeOptions, Message};
use tokio_stream::StreamExt;

// Simple query
let mut stream = query("Hello Claude", None).await?;
while let Some(message) = stream.next().await {
    match message {
        Message::Assistant(msg) => {
            for block in &msg.content {
                if let ContentBlock::Text(text_block) = block {
                    println!("Claude: {}", text_block.text);
                }
            }
        }
        _ => {}
    }
}

// With options
let options = ClaudeCodeOptions {
    system_prompt: Some("You are a helpful assistant".to_string()),
    max_turns: Some(1),
    ..Default::default()
};

let mut stream = query("Tell me a joke", Some(options)).await?;
while let Some(message) = stream.next().await {
    println!("{:?}", message);
}

Using Tools

let options = ClaudeCodeOptions {
    allowed_tools: vec!["Read".to_string(), "Write".to_string()],
    permission_mode: Some(PermissionMode::AcceptEdits),
    ..Default::default()
};

let mut stream = query("Create a hello.rs file", Some(options)).await?;
while let Some(message) = stream.next().await {
    // Process tool use and results
}

Working Directory

use std::path::PathBuf;

let options = ClaudeCodeOptions {
    cwd: Some(PathBuf::from("/path/to/project")),
    ..Default::default()
};

Logging

The Claude Code SDK provides comprehensive structured logging using the tracing ecosystem. This helps with debugging, monitoring, and understanding SDK operations.

Quick Setup

use claude_code_sdk;

#[tokio::main]
async fn main() -> Result<(), Box<dyn std::error::Error>> {
    // Initialize default logging
    claude_code_sdk::init_tracing();
    
    // Your code here...
    Ok(())
}

Environment Variables

Control logging levels with the RUST_LOG environment variable:

# Show info and above (default)
RUST_LOG=claude_code_sdk=info cargo run

# Show debug messages (recommended for development)
RUST_LOG=claude_code_sdk=debug cargo run

# Show all messages including trace
RUST_LOG=claude_code_sdk=trace cargo run

# Show only errors
RUST_LOG=claude_code_sdk=error cargo run

# Multiple modules
RUST_LOG=claude_code_sdk=debug,tokio=info cargo run

Custom Logging Setup

For more control over logging, set up tracing-subscriber directly:

use tracing_subscriber::{layer::SubscriberExt, util::SubscriberInitExt};

// JSON structured logging
tracing_subscriber::registry()
    .with(tracing_subscriber::EnvFilter::new("claude_code_sdk=debug"))
    .with(tracing_subscriber::fmt::layer().json())
    .init();

// Custom formatting
tracing_subscriber::registry()
    .with(tracing_subscriber::EnvFilter::new("claude_code_sdk=debug"))
    .with(
        tracing_subscriber::fmt::layer()
            .with_target(true)
            .with_thread_ids(true)
            .with_file(true)
            .with_line_number(true)
            .pretty()
    )
    .init();

// File logging
let file = std::fs::OpenOptions::new()
    .create(true)
    .write(true)
    .truncate(true)
    .open("claude_sdk.log")?;

tracing_subscriber::registry()
    .with(tracing_subscriber::EnvFilter::new("claude_code_sdk=trace"))
    .with(tracing_subscriber::fmt::layer().with_writer(file))
    .init();

What Gets Logged

The SDK logs various operations at different levels:

• ERROR: CLI not found, connection failures, JSON decode errors
• WARN: Process termination issues, unexpected states
• INFO: Query start/completion, connection events, major operations
• DEBUG: Message processing, command building, subprocess management
• TRACE: Individual message parsing, detailed state changes

Log Fields

Structured logs include useful fields:

{
  "timestamp": "2024-01-01T12:00:00.000Z",
  "level": "INFO",
  "target": "claude_code_sdk::client",
  "message": "Processing query through transport",
  "fields": {
    "prompt_length": 25,
    "has_options": true,
    "system_prompt": "You are helpful",
    "allowed_tools": 2,
    "permission_mode": "AcceptEdits"
  }
}

Performance Monitoring

Enable performance monitoring with tracing spans:

use tracing::{info, Instrument};

async fn my_function() -> Result<(), Box<dyn std::error::Error>> {
    let span = tracing::info_span!("my_operation", operation_type = "query");
    
    async move {
        let start = std::time::Instant::now();
        
        // Your Claude SDK operations here
        let mut stream = query("Hello", None).await?;
        // ... process stream
        
        info!(duration_ms = start.elapsed().as_millis(), "Operation completed");
        Ok(())
    }
    .instrument(span)
    .await
}

Safety & Long Text Handling

The Claude Code SDK includes comprehensive safety mechanisms to handle long text responses safely and prevent common issues like memory exhaustion and performance degradation.

Safety Limits

Configure safety limits to protect against resource exhaustion:

use claude_code_sdk::{SafetyLimits, SafetyError, ClaudeSDKError};

// Default limits (balanced)
let default_limits = SafetyLimits::default();

// Conservative limits (memory-constrained environments)
let conservative = SafetyLimits::conservative();

// Generous limits (high-memory servers)
let generous = SafetyLimits::generous();

// Custom limits
let custom = SafetyLimits {
    max_line_size: 5 * 1024 * 1024,      // 5MB per line
    max_text_block_size: 2 * 1024 * 1024, // 2MB per text block
    max_buffer_size: 20 * 1024 * 1024,   // 20MB total buffer
    max_buffered_messages: 50,            // 50 messages max
    json_parse_timeout_ms: 3000,          // 3 second timeout
    max_log_preview_chars: 150,           // 150 char preview
};

Protected Operations

The SDK automatically protects against:

• Memory Exhaustion: Line size and text block limits prevent runaway memory usage
• JSON Parsing Timeouts: Large JSON objects are monitored for parsing time
• Log Explosion: Automatic truncation of log messages with size indicators
• Buffer Overflow: Channel size limits prevent message queue buildup

Error Handling

Safety violations are reported through structured errors:

match query("Generate a very long response", None).await {
    Ok(stream) => {
        // Process normally
    }
    Err(ClaudeSDKError::Safety(SafetyError::LineTooLarge { actual, limit })) => {
        println!("Line too large: {} bytes (limit: {})", actual, limit);
    }
    Err(ClaudeSDKError::Safety(SafetyError::TextBlockTooLarge { actual, limit })) => {
        println!("Text block too large: {} bytes (limit: {})", actual, limit);
    }
    Err(ClaudeSDKError::Safety(SafetyError::ParseTimeout { timeout_ms })) => {
        println!("JSON parsing timeout after {}ms", timeout_ms);
    }
    Err(other) => {
        println!("Other error: {}", other);
    }
}

Risk Mitigation

Monitoring

The SDK provides detailed monitoring of resource usage:

use tracing::{info, warn};

// Automatic logging of resource usage
// INFO: Text block size monitoring
// WARN: Approaching safety limits  
// ERROR: Safety limit violations

// Example log output:
// INFO text_length=1048576 limit=5242880 "Large text block detected"
// WARN "Text block is approaching size limit"
// ERROR actual=10485760 limit=5242880 "Line exceeds safety limit"

API Reference

query(prompt: &str, options: Option<ClaudeCodeOptions>)

Main async function for querying Claude.

Parameters:

• prompt: The prompt to send to Claude
• options: Optional configuration

Returns: Result<Pin<Box<dyn Stream<Item = Message>>>, ClaudeSDKError>

Types

See src/types.rs for complete type definitions:

• ClaudeCodeOptions - Configuration options
• Message variants - Assistant, User, System, Result
• ContentBlock variants - Text, ToolUse, ToolResult

Error Handling

use claude_code_sdk::{
    ClaudeSDKError,      // Base error
    CLINotFoundError,    // Claude Code not installed
    CLIConnectionError,  // Connection issues
    ProcessError,        // Process failed
    CLIJSONDecodeError,  // JSON parsing issues
};

match query("Hello", None).await {
    Ok(mut stream) => {
        // Process messages
    }
    Err(ClaudeSDKError::CLINotFound(e)) => {
        println!("Please install Claude Code: {}", e);
    }
    Err(ClaudeSDKError::Process(e)) => {
        println!("Process failed with exit code: {:?}", e.exit_code);
    }
    Err(e) => {
        println!("Error: {}", e);
    }
}

Available Tools

See the Claude Code documentation for a complete list of available tools.

Examples

• examples/quick_start.rs - Complete working example with logging
• examples/logging_demo.rs - Advanced logging configurations
• examples/safety_demo.rs - Safety mechanisms and long text handling

Running Examples

# Basic example with default logging
cargo run --example quick_start

# Advanced logging demo with JSON output
cargo run --example logging_demo -- --json

# File logging
cargo run --example logging_demo -- --file logs/claude.log

# Performance monitoring
RUST_LOG=claude_code_sdk=debug cargo run --example logging_demo -- --perf

# Safety demonstration
cargo run --example safety_demo -- --simulate-large

License

MIT