Claude Agent SDK for Rust

December 23, 2025 · View on GitHub

Rust SDK for building AI agents powered by Claude Code. Mirrors the TypeScript Claude Agent SDK with idiomatic Rust patterns.

Installation

Prerequisites:

  • Rust 1.85.0+ (edition 2024)
  • Node.js
  • Claude Code 2.0.75+: npm install -g @anthropic-ai/claude-code
[dependencies]
anthropic-agent-sdk = "0.2"
tokio = { version = "1", features = ["rt-multi-thread", "macros"] }

Quick Start

use anthropic_agent_sdk::{query, Message, ContentBlock, StreamExt};

#[tokio::main]
async fn main() -> Result<(), Box<dyn std::error::Error>> {
    let stream = query("What is 2 + 2?", None).await?;
    let mut stream = Box::pin(stream);

    while let Some(message) = stream.next().await {
        if let Message::Assistant { message, .. } = message? {
            for block in &message.content {
                if let ContentBlock::Text { text } = block {
                    println!("{}", text);
                }
            }
        }
    }
    Ok(())
}

Examples

# Core functionality
cargo run --example simple_query
cargo run --example convenience_methods
cargo run --example bidirectional_demo
cargo run --example message_queue_demo
cargo run --example session_binding_demo
cargo run --example interactive_client

# Hooks and permissions
cargo run --example hooks_demo
cargo run --example permissions_demo
cargo run --example hooks_lifecycle_test

# Introspection and runtime
cargo run --example introspection_demo
cargo run --example result_fields_demo
cargo run --example runtime_setters_demo

# Security
cargo run --example security_demo

# OAuth authentication
cargo run --example oauth_demo
cargo run --example oauth_demo -- status
cargo run --example oauth_demo -- logout

# Structured output
cargo run --example structured_output_demo

# MCP (requires --features rmcp for mcp_server)
cargo run --example mcp_integration
cargo run --example mcp_server --features rmcp

TypeScript SDK Parity

Hook Events

EventTypeScriptRust SDK
PreToolUse✓✓
PostToolUse✓✓
PostToolUseFailure✓✓
Notification✓✓
UserPromptSubmit✓✓
SessionStart✓✓
SessionEnd✓✓
Stop✓✓
SubagentStart✓✓
SubagentStop✓✓
PreCompact✓✓
PermissionRequest✓✓

Query/Client Methods

MethodTypeScriptRust SDK
interrupt()✓✓
setPermissionMode()✓✓
setModel()✓✓
setMaxThinkingTokens()✓✓
supportedCommands()✓✓
supportedModels()✓✓
mcpServerStatus()✓✓
accountInfo()✓✓

Options

OptionTypeScriptRust SDK
allowedTools✓✓
disallowedTools✓✓
systemPrompt✓✓
mcpServers✓✓
permissionMode✓✓
canUseTool✓✓
hooks✓✓
agents✓✓
maxTurns✓✓
model✓✓
cwd✓✓
env✓✓
resume✓✓
forkSession✓✓
settingSources✓✓
maxBudgetUsd✓✓
maxThinkingTokens✓✓
fallbackModel✓✓
outputFormat✓✓
sandbox✓✓
plugins✓✓
betas✓✓
strictMcpConfig✓✓
resumeSessionAt✓✓
allowDangerouslySkipPermissions✓✓
pathToClaudeCodeExecutable✓✓
stderr✓✓
tools (preset)✓✓
enableFileCheckpointing✓✓
sessionId✓✓

MCP Server Types

TypeTypeScriptRust SDK
stdio✓✓
sse✓✓
http✓✓
sdk (in-process)✓✓

Result Message Fields

FieldTypeScriptRust SDK
modelUsage✓✓
permission_denials✓✓
structured_output✓✓
errors✓✓

Development

cargo build
cargo test
cargo clippy
cargo doc --open

Security

The SDK implements strict security measures:

  • Environment variables: Dangerous vars (LD_PRELOAD, PATH, etc.) cause errors
  • CLI arguments: Only allowlisted flags permitted, others rejected
  • Session binding: Auto-binds on first Result, validates on send
  • Buffer limits: Configurable max buffer size (default 1MB)
  • 100% safe Rust: No unsafe code

See SECURITY.md for full documentation.

OAuth Authentication

For Claude Max/Pro subscribers, authenticate without API keys:

use anthropic_agent_sdk::auth::OAuthClient;

let client = OAuthClient::new()?;
let token = client.authenticate().await?;
// Token is cached in platform-specific config directory

See oauth_demo example for full usage including status check and logout.

Documentation

License

MIT