07 - SDKs: Python, TypeScript, and CLI¶

Anthropic ships first-party SDKs for Python and TypeScript and a CLI for quick experimentation. The exam expects fluency in idiomatic SDK usage, including streaming helpers, async clients, and the Bedrock / Vertex sub-clients.

Python SDK¶

Install:

pip install anthropic

Sync Client¶

from anthropic import Anthropic

client = Anthropic()  # reads ANTHROPIC_API_KEY from env
msg = client.messages.create(
    model="claude-sonnet-4-6",
    max_tokens=1024,
    messages=[{"role": "user", "content": "Hi"}],
)
print(msg.content[0].text)

Async Client¶

from anthropic import AsyncAnthropic
import asyncio

aclient = AsyncAnthropic()

async def main():
    msg = await aclient.messages.create(
        model="claude-sonnet-4-6",
        max_tokens=1024,
        messages=[{"role": "user", "content": "Hi"}],
    )
    print(msg.content[0].text)

asyncio.run(main())

Use async for fan-out (parallel requests, agent workers).

Streaming Helpers¶

with client.messages.stream(
    model="claude-sonnet-4-6",
    max_tokens=1024,
    messages=[{"role": "user", "content": "Tell me a story"}],
) as stream:
    for text in stream.text_stream:
        print(text, end="", flush=True)
    final = stream.get_final_message()

Async streaming:

async with aclient.messages.stream(...) as stream:
    async for text in stream.text_stream:
        print(text, end="", flush=True)

For tool_use streaming, iterate over events directly via for event in stream:.

Configuration Options¶

client = Anthropic(
    api_key="...",          # or env
    base_url="...",         # for proxies
    timeout=30.0,
    max_retries=2,          # SDK-level retries
    default_headers={...},
)

Per-Request Options¶

client.with_options(timeout=60.0, max_retries=0).messages.create(...)

Error Types¶

from anthropic import (
    APIError, APIConnectionError, APITimeoutError,
    BadRequestError, AuthenticationError, PermissionDeniedError,
    NotFoundError, RateLimitError, InternalServerError,
)

Bedrock Client¶

from anthropic import AnthropicBedrock

client = AnthropicBedrock(
    aws_region="us-east-1",
    # AWS credentials via env, profile, or IAM role
)
msg = client.messages.create(
    model="anthropic.claude-sonnet-4-6-YYYYMMDD-v1:0",
    max_tokens=1024,
    messages=[...],
)

Vertex Client¶

from anthropic import AnthropicVertex

client = AnthropicVertex(region="us-east5", project_id="my-gcp-project")
msg = client.messages.create(
    model="claude-sonnet-4-6@YYYYMMDD",
    max_tokens=1024,
    messages=[...],
)

The Messages interface is identical across all three clients.

TypeScript SDK¶

Install:

npm install @anthropic-ai/sdk

Basic Usage¶

import Anthropic from "@anthropic-ai/sdk";

const client = new Anthropic();  // reads ANTHROPIC_API_KEY

const msg = await client.messages.create({
  model: "claude-sonnet-4-6",
  max_tokens: 1024,
  messages: [{ role: "user", content: "Hi" }],
});

console.log(msg.content[0]);

Streaming¶

const stream = client.messages.stream({
  model: "claude-sonnet-4-6",
  max_tokens: 1024,
  messages: [{ role: "user", content: "Tell me a story" }],
});

stream.on("text", (delta) => process.stdout.write(delta));

const final = await stream.finalMessage();

Or async iteration:

for await (const event of stream) {
  if (event.type === "content_block_delta") { ... }
}

Configuration¶

const client = new Anthropic({
  apiKey: process.env.ANTHROPIC_API_KEY,
  baseURL: "...",
  timeout: 30_000,
  maxRetries: 2,
});

Error Handling¶

import { APIError, RateLimitError } from "@anthropic-ai/sdk";

try {
  await client.messages.create({...});
} catch (err) {
  if (err instanceof RateLimitError) { ... }
  else if (err instanceof APIError) { ... }
  else throw err;
}

Bedrock and Vertex (TS)¶

import { AnthropicBedrock } from "@anthropic-ai/bedrock-sdk";
import { AnthropicVertex } from "@anthropic-ai/vertex-sdk";

const bedrock = new AnthropicBedrock({ awsRegion: "us-east-1" });
const vertex = new AnthropicVertex({ region: "us-east5", projectId: "..." });

(Package names may vary; check current docs.)

CLI Tools¶

The Python SDK does not ship a one-line anthropic CLI by default, but claude-agent-sdk and Claude Code provide CLI surfaces. For ad-hoc API testing, use curl:

curl https://api.anthropic.com/v1/messages \
  -H "x-api-key: $ANTHROPIC_API_KEY" \
  -H "anthropic-version: 2023-06-01" \
  -H "content-type: application/json" \
  -d '{"model": "claude-sonnet-4-6", "max_tokens": 256, "messages": [{"role": "user", "content": "Hi"}]}'

For developer workflows, Claude Code is the official CLI agent.

Idiomatic Patterns¶

Reuse the Client¶

Construct one client per process and reuse it. Connection pooling, retries, and headers are stable across calls.

Type Safety¶

Both SDKs ship full typings. In Python, the response objects are dataclass-like; access fields by attribute. In TypeScript, types flow from the imports.

Pagination¶

Some endpoints (Files list, Batches list) return paginated results. The SDKs expose iterators:

for f in client.files.list():
    print(f.id)

Webhook Verification¶

If you receive Anthropic webhooks (e.g., for batch completion), verify signatures per docs. The SDKs may expose helpers.

Test Doubles¶

For tests, mock the client at the boundary. Avoid mocking inside the SDK. Use replay fixtures for streaming behavior.

Cross-SDK Choice¶

Both SDKs cover the same Messages surface. Pick by ecosystem.

SDK Versioning¶

The SDKs add features as Anthropic ships them. Stay reasonably current; old SDK versions may not expose new features (memory tool, citations, code execution). Pin a known-good minor version in your manifest, upgrade deliberately.

Common Pitfalls¶

• Constructing a new client per request (defeats connection pooling)
• Catching all exceptions broadly
• Mocking too deep into the SDK
• Hardcoding model IDs in source
• Mixing Bedrock model IDs with first-party model IDs
• Leaving streams unconsumed

Exam Focus¶

• Which client class to use for first-party / Bedrock / Vertex
• Streaming helpers and how to assemble final messages
• Async vs sync client choice
• Error class hierarchy
• Per-request option overrides
• Idiomatic agent loop construction with the SDK