Everyinfra Cli

---
name: everyinfra-cli
description: Fetch public data from 100+ platforms (TikTok, Amazon, Instagram, Xiaohongshu, Google Maps and more), run web searches, solve CAPTCHAs and call an OpenAI-compatible AI API with the `everyinfra` command-line tool and one EveryInfra API key. Use it when a task needs platform data, search results or a CAPTCHA solution and a terminal is available.
---

# EveryInfra CLI

`everyinfra` is a thin command-line client for the EveryInfra API: platform data, web search, CAPTCHA
solving and the AI API, all paid from one wallet with one key.

Install: `pipx install everyinfra` (or `uv tool install everyinfra`, or `pip install everyinfra`).
Check the setup with `everyinfra doctor`.

## CLI or MCP

- Use the CLI when you have a shell: output goes to files and pipes, and large results do not fill your context.
- Use the MCP server (`https://api.everyinfra.com/mcp`) in chat apps without a terminal.
  `everyinfra mcp-config --client claude|cursor|codex|vscode` prints the config.

## The key

The CLI reads `--key`, then `EVERYINFRA_API_KEY`, then the key saved by `everyinfra login`.
Never print the key, put it in a file you share, or include it in a support ticket.
The catalog, `describe`, `captcha types`, `whats-new` and `--dry-run` work without a key.

## Standard flow

1. Find the capability (free): `everyinfra catalog --find "tiktok comments"`.
   Search tools: `everyinfra catalog --line search`. CAPTCHA types: `everyinfra captcha types`.
2. Read its contract (free): `everyinfra describe tiktok/comments`. It lists the parameters, allowed values,
   returned fields, the price and how paging works.
3. Preview the call (free): `everyinfra data tiktok comments -p url=... --limit 50 --dry-run`.
   It prints the exact request and the estimated cost. Nothing is sent.
4. Try a small real call: `--limit 5`. Check the result before you spend more.
5. Run the batch with a budget: `--max-cost-usd 2 --output result.json`. The CLI does not send a call whose
   estimate is over the budget.

Ask the user before a batch that costs more than they expect. Every paid command prints the charge and the
balance on stderr. `everyinfra account` shows the balance; `everyinfra billing` shows the spend.

## Parameters

- `-p key=value` is converted to the type in the catalog (integer, boolean, array and so on).
- `-p key:=JSON` passes raw JSON: `-p urls:='["https://a","https://b"]'`.
- Repeat `-p` for an array: `-p urls=https://a -p urls=https://b`. Arrays of targets are charged per target.
- Slow capabilities or large limits: add `--async --wait`, or `--async` and later `everyinfra job JOB_ID --wait`.
- Paged capabilities: `--pages 3` follows `next_page_token`. Each page is a separate paid call.

## Output

- Data goes to stdout and messages go to stderr.
- Output is JSON when piped or with `--json`. Use `--jsonl` for one result per line and `--fields a,b` for
  only some fields.
- For large results use `--output FILE`. Then stdout gets only a short summary.

## Exit codes

| Code | Meaning | What to do |
|---|---|---|
| 0 | OK | |
| 2 | Bad request (400, 422) | Fix the parameters. The error JSON on stderr has `param`, `did_you_mean` and `allowed`. |
| 3 | Auth or permission (401, 403) | Fix the key, or ask the user to top up (`ai_topup_required`) or enable a key permission. |
| 4 | Balance too low (402) | Ask the user to top up. Do not retry. |
| 5 | Rate limited (429) | Wait `retry_after` seconds from the error JSON, then retry. |
| 6 | Temporary failure (503) | Retry later. Failed calls are not charged. |
| 7 | Outcome unknown | Do not retry. The call may have run and been charged. Check it with `everyinfra requests --id REQUEST_ID` or `everyinfra job JOB_ID`. |
| 1 | Other | Read the error JSON. |

## Errors and support

Each error on stderr has the API's shape: `{"error": {"code", "message", "request_id", ...}}`. Keep the
`request_id`. When the error says the problem is on EveryInfra's side, draft a ticket with
`everyinfra support draft --subject ... --description ... --request-id ...`, show it to the user, and open
it with `everyinfra support open ... --yes` only after they agree.

## AI API

Accounts that have topped up can use the AI API at no charge: `everyinfra ai chat "..."` for one reply,
`everyinfra ai map --input rows.jsonl --instruction "..." --output out.jsonl` to label, translate or extract
fields row by row. `ai map` waits on rate limits and keeps the input order. Add `--resume` to continue a run.

Guide for agents: https://everyinfra.com/llms.txt