EveryInfra API and MCP documentation

Authentication

Send the API key as a Bearer token. Authentication runs before parameter validation, so a request without a valid key receives 401 rather than parameter feedback. Keep keys out of browser code, logs, and public repositories.

• Create separate keys for separate environments or workloads.
• Rotate a key from the console when its scope or exposure changes.
• Use the console's restrictions when a key should be limited to specific product lines, capabilities, or IP ranges.
• Use the same key with GET /api/v1/account to check your balance (paid, bonus, and total), this month's top-up discount, the per-minute request limit, and the last 24 hours of usage. Amounts are in USD. The call is read-only, free, and does not count toward the per-minute limit.
• In that response, balance.total_usd is paid plus bonus balance, and calls draw on the bonus balance first. bonus_expiring lists time-limited bonus amounts that have not expired yet. rate_limit.requests_per_minute is shared by every key on the account, and usage_24h can lag by about a minute.
• For account-wide history (daily usage, recent calls, billing, the wallet ledger), the same key can read the console API, for example GET /portal/api/requests/{request_id}. The key needs account read access; these reads are free and limited to 60 per minute per account. The full list is in the Account section of the OpenAPI spec.
• To let an AI change the account too, grant write scopes under API Keys → Restrict access: manage keys, change settings, and open or reply to tickets. All three are off by default and only you can turn them on, signed in to the console. Through a key, at most 20 keys and 10 tickets a day can be created; viewing or rotating full keys, changing permissions or IP allowlists, re-enabling keys, and payments always stay in the console. The Account section of the OpenAPI spec lists the endpoints, for example POST /portal/api/keys/{key_id}/disable.
• To avoid polling, add an HTTPS endpoint under Settings → Webhooks in the console. We send a signed POST when the balance falls to your threshold or an async job finishes. Verify X-EveryInfra-Signature, an HMAC-SHA256 of the timestamp, a period and the raw body made with your signing secret; reject timestamps more than 5 minutes off and de-duplicate on the event ID, since failed deliveries are retried with backoff, up to 7 attempts.

curl https://api.everyinfra.com/api/v1/account \
  -H "Authorization: Bearer YOUR_API_KEY"

AI API (free after a top-up)

The AI API is OpenAI-compatible chat/completions. Any account that has topped up calls it at no charge, without using its balance, to clean, deduplicate, classify, extract, and summarize the data it collects; the prompt sets the output format.

• With the OpenAI SDK, set base_url to https://api.everyinfra.com/api/v1 and send your EveryInfra API key in the Authorization: Bearer header; a ?key= in the URL is rejected.
• The chat endpoint is POST /api/v1/chat/completions, with OpenAI's request and response format.
• There are two models: gemini-3.6-flash, the default, and the lighter gemini-flash-lite. GET /api/v1/models returns the same list and needs no key. A misspelled model returns 422 with the likely correct name.
• Streaming is supported: send stream: true to receive SSE that ends with data: [DONE], and add stream_options.include_usage for token usage.
• messages may total up to 45,000 characters; larger inputs return 422 input_too_large, so split them into smaller requests.

from openai import OpenAI
import os

client = OpenAI(base_url="https://api.everyinfra.com/api/v1",
                api_key=os.environ["EVERYINFRA_API_KEY"])
reply = client.chat.completions.create(
    model="gemini-3.6-flash",
    messages=[{"role": "user", "content": "Return JSON only: label this review's sentiment, topic and intent."}],
)
print(reply.choices[0].message.content)

Limits by total top-ups, across all keys on the account

Trial credit and credit codes do not count toward total top-ups; an account that has never topped up gets 403 ai_topup_required. Tiers are ceilings, not guaranteed throughput.

The AI API is for organizing the data you collect with EveryInfra. If it is used for something else, or the calls look clearly abnormal, we may pause the AI API for that account; the account itself, its other APIs, and its balance are not affected.

Understand charges and balance adjustments

Check the public catalog for the price and billing unit of the capability you need. Each response records the amount charged. For products billed on successful delivery, failed or empty results are not charged.

• Do not convert internal wallet units into currency in your client.
• Display the currency and amounts returned by the API, and use them to reconcile charges.
• When a delivery-based operation does not complete, no final charge remains and any reserved amount returns to the account balance automatically.
• Each paid response's billing object includes charged_usd, the amount charged for that call in USD (0 when nothing was charged, null while the outcome is still unknown), and balance_after_usd, the wallet balance once the call settles (null if it cannot be read). Watch balance_after_usd during a batch instead of polling the account.
• When a failed call's reserved amount has been returned, the error body's billing also carries balance_after_usd.