SDKs and frameworks

Anthropic SDK

Use the official Anthropic SDKs for Python and TypeScript with XField: messages, streaming, thinking, tools, and how errors map to SDK classes.

XField serves the Anthropic Messages API, so the official anthropic SDKs work with a different base URL and your key. Thinking comes back as real thinking blocks and tools use the native tool_use format.

Install and configure

Shell

pip install anthropic              # Python
npm install @anthropic-ai/sdk      # TypeScript / Node

client.py

Python

import os
import anthropic
 
client = anthropic.Anthropic(
    base_url="https://xfield.cloud.brutal.ai/api",
    api_key=os.environ["XFIELD_API_KEY"],
)

Use https://xfield.cloud.brutal.ai/api, not the /v1 URL: the SDK appends /v1/messages itself. The key goes out as x-api-key.

Messages

Models that think by default return a thinking block before the text, so pick the text block by type rather than taking content[0].

Python

msg = client.messages.create(
    model="deepseek-v4.1-flash-1017-uncensored",
    max_tokens=2048,
    system="You are a terse security reviewer.",
    messages=[{"role": "user", "content": "Explain TOCTOU in two sentences."}],
)
print(next(b.text for b in msg.content if b.type == "text"))

TypeScript

const msg = await client.messages.create({
  model: "deepseek-v4.1-flash-1017-uncensored",
  max_tokens: 2048,
  system: "You are a terse security reviewer.",
  messages: [{ role: "user", content: "Explain TOCTOU in two sentences." }],
});
const text = msg.content.find((b) => b.type === "text");
console.log(text?.type === "text" ? text.text : "");

Streaming

Python

with client.messages.stream(
    model="deepseek-v4.1-flash-1017-uncensored",
    max_tokens=2048,
    messages=[{"role": "user", "content": "Draft a lab setup for testing EDR evasion detections."}],
) as stream:
    for text in stream.text_stream:
        print(text, end="", flush=True)
    final = stream.get_final_message()
print("\n", final.usage)

TypeScript

const stream = client.messages.stream({
  model: "deepseek-v4.1-flash-1017-uncensored",
  max_tokens: 2048,
  messages: [{ role: "user", content: "Draft a lab setup for testing EDR evasion detections." }],
});
stream.on("text", (t) => process.stdout.write(t));
const final = await stream.finalMessage();
console.log("\n", final.usage);

Thinking

Control thinking with thinking (enabled, adaptive or disabled) and depth with output_config.effort. budget_tokens is accepted for compatibility but is not enforced as a separate budget; use effort instead.

Python

msg = client.messages.create(
    model="glm-5.3-uncensored",
    max_tokens=16000,
    thinking={"type": "adaptive"},
    output_config={"effort": "high"},
    messages=[{"role": "user", "content": "Rank these three findings by exploitability and say why."}],
)
for block in msg.content:
    if block.type == "thinking":
        print("[thinking]", block.thinking[:200])
    elif block.type == "text":
        print(block.text)

If your SDK version rejects output_config as an unknown argument, pass it through extra_body (Python) or upgrade. Accepted effort words are low, medium, high, xhigh and max; each model maps them to the levels it has.

Tools

Python

tools = [{
    "name": "run_query",
    "description": "Run a read-only SQL query against the test database.",
    "input_schema": {
        "type": "object",
        "properties": {"sql": {"type": "string"}},
        "required": ["sql"],
    },
}]
 
msg = client.messages.create(
    model="deepseek-v4.1-flash-1017-uncensored",
    max_tokens=2048,
    tools=tools,
    messages=[{"role": "user", "content": "How many users have no password set?"}],
)
if msg.stop_reason == "tool_use":
    call = next(b for b in msg.content if b.type == "tool_use")
    print(call.name, call.input)

Send the whole assistant content back, then a user message with a tool_result block carrying tool_use_id=call.id, and call again. Keep the thinking blocks in the history when you do.

Errors

Errors on /messages use Anthropic's shape, so the SDK raises its usual classes. Names are the same in both SDKs.

Status	SDK class	What it means here
400	`BadRequestError`	Malformed request or unknown model id
401	`AuthenticationError`	Key missing, unknown, disabled or expired
403	`PermissionDeniedError`	Account disabled, or content policy not yet confirmed
415	`APIStatusError`	Body was not JSON (the SDK always sends JSON)
429	`RateLimitError`	Rate limited; back off and retry
5xx	`InternalServerError`	Upstream trouble; a 503 with `Retry-After` is safe to retry
none	`APIConnectionError`, `APITimeoutError`	Network failure or timeout

The SDKs already retry 429s, 5xx responses and connection errors twice with backoff. Raise that with max_retries (Python) or maxRetries (TypeScript) on the client, and set timeout generously for long thinking runs. See limits and errors for the full list.