Messages (Anthropic-compatible)

You're Claude Code, working in my repository.

Task
Call the two Messages endpoints from my app in the Anthropic wire format: POST /v1/messages to generate (model, max_tokens, messages, optional system, tools and thinking) and POST /v1/messages/count_tokens to measure a request first, reading content blocks, stop_reason and usage from the reply.

Read this first
Meta's docs page "Messages (Anthropic-compatible)" as Markdown: https://dev.meta.ai/docs/api-reference/messages.md
It's the source of truth for request and response shapes. The same page for people: https://dev.meta.ai/docs/api-reference/messages

Setup
- Base URL: https://api.meta.ai/v1 for HTTP calls (POST https://api.meta.ai/v1/messages); the Anthropic SDK takes https://api.meta.ai
- Auth: send Authorization: Bearer $MODEL_API_KEY, reading MODEL_API_KEY from the environment. Never hard-code, print or commit the key.
- Model: muse-spark-1.3-contributor (Contributor mode: Meta may use what you send to train future models; drop -contributor for Standard mode)
- Python: the Anthropic SDK (pip install anthropic) with base_url="https://api.meta.ai" (no /v1) and auth_token=os.environ["MODEL_API_KEY"], or plain requests to https://api.meta.ai/v1/messages.

Facts to get right
- Auth is Authorization: Bearer $MODEL_API_KEY (x-api-key is accepted too); anthropic-version is accepted for SDK compatibility.
- Supported tools are developer-defined custom tools and the built-in web_search.
- count_tokens takes the same body minus max_tokens and returns {"input_tokens": n}.
- Rate limits and billing are the same as the other endpoints.

Starter code (Python)
import os

from anthropic import Anthropic

client = Anthropic(base_url="https://api.meta.ai", auth_token=os.environ["MODEL_API_KEY"])
MODEL = "muse-spark-1.3-contributor"

body = {"model": MODEL, "messages": [{"role": "user", "content": "What is the capital of France?"}]}
print(client.messages.count_tokens(**body).input_tokens)
message = client.messages.create(max_tokens=1024, **body)
print(message.stop_reason, [b.type for b in message.content], message.usage)

Done when
- A message call prints its text blocks and usage; a count_tokens call prints input_tokens for the same body.
- The key comes from MODEL_API_KEY and never appears in code, logs or commits.
- API errors (4xx, 429) surface their message instead of failing silently.