Count input tokens

You're Claude Code, working in my repository.

Task
Before my app sends a long Responses request, measure it: post the same body to POST /v1/responses/input_tokens, check that input_tokens plus max_output_tokens fits the model's context window, and drop the oldest whole turns until it does.

Read this first
Meta's docs page "Count input tokens" as Markdown: https://dev.meta.ai/docs/api-reference/responses/count-input-tokens.md
It's the source of truth for request and response shapes. The same page for people: https://dev.meta.ai/docs/api-reference/responses/count-input-tokens

Setup
- Base URL: https://api.meta.ai/v1
- 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 OpenAI SDK (pip install openai) with base_url="https://api.meta.ai/v1" and api_key=os.environ["MODEL_API_KEY"] works, or plain requests.

Facts to get right
- The body is TokenCountsBody: the same fields as a create (model, input, instructions, tools, previous_response_id, ...). Nothing is generated or stored.
- The reply is {"object": "response.input_tokens", "input_tokens": n}: the fully rendered input, with resolved previous_response_id history, instructions, tools, media and server scaffolding.
- It's a context-size measurement, not billing: usage.input_tokens on the real call can differ.
- Muse Spark and Muse Wonder have a 1,048,576-token context window, which input_tokens plus max_output_tokens must fit.
- Conversation resources aren't supported. An over-context create is a 400: truncation only takes "disabled".
- When trimming encrypted reasoning items, keep or drop each one whole.

Starter code (Python)
import os

import requests

body = {
    "model": "muse-spark-1.3-contributor",
    "instructions": "Answer in one sentence.",
    "input": "What is the capital of France?",
}
response = requests.post(
    "https://api.meta.ai/v1/responses/input_tokens",
    headers={"Authorization": f"Bearer {os.environ['MODEL_API_KEY']}"},
    json=body,
)
response.raise_for_status()
print(response.json()["input_tokens"])

Done when
- The app prints the count for a request, and a request trimmed to fit still runs.
- 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.