
Most captcha task types are a two-step client. You create a task, store the taskId, and poll until the worker is done. AntiAwsWafTask is the type Kaptra documents as synchronous. /createTask returns status=ready and the token in the same response. If your wrapper always polls, this is the call that should skip that branch.
This is a practical walkthrough of that call. It is not the guide for the AWS WAF image grid. That type is AwsWafClassification, it is asynchronous, and it returns tile indices instead of a token.
The task in one glance
- Type:
AntiAwsWafTask - Required:
websiteURL, the public URL of the challenged page - Optional:
referrer, which defaults fromwebsiteURL - Optional, as a set:
awsKey,awsIv,awsContext. All three, or none - Not a parameter:
proxy - Execution: synchronous.
/createTaskreturns the result. Do not poll - Result:
solution.token, shaped likeaws-waf-token=... - Cost and speed: $0.60 per 1,000 successful tasks, average under 3 seconds, billed per success only
Why there is no poll loop
The introduction is explicit. Some types, today AntiAwsWafTask, return status=ready from /createTask. Everything else returns status=processing and a taskId, and you poll /getTaskResult every 1 to 2 seconds, at most 120 times per task.
A shared client can keep the poll branch for the other types. For this type, the branch is dead code. There is no taskId to query, and querying one you invented will not make the call more correct.
The HTTP status is still always 200. Read errorId before you read solution. It is 0 or 1, and the reason is errorCode.
Working Python implementation
import requests
BASE_URL = "https://api.kaptra.ai"
API_KEY = "YOUR_KAPTRA_API_KEY"
def solve_aws_waf(website_url: str) -> dict:
# synchronous: createTask returns the result, no polling
resp = requests.post(
f"{BASE_URL}/createTask",
json={
"clientKey": API_KEY,
"task": {
"type": "AntiAwsWafTask",
"websiteURL": website_url,
},
},
).json()
if resp.get("errorId"):
raise RuntimeError(f"createTask failed: {resp.get('errorCode')}")
if resp.get("status") != "ready":
raise RuntimeError(f"unexpected status: {resp.get('status')}")
return resp["solution"] # {"token": "aws-waf-token=..."}
if __name__ == "__main__":
page = "https://www.example.com/page"
solution = solve_aws_waf(page)
token = solution["token"]
print(token)
# retry the page with the token as a cookie
value = token.split("=", 1)[1]
session = requests.Session()
session.cookies.set("aws-waf-token", value)
page_resp = session.get(page)
print(page_resp.status_code)
The retry is the second half of the function, not a later session. Split the token on the first =, set a cookie named aws-waf-token, and request the same page again. The documented solution has no expiry field and no user agent, so there is nothing else to copy across.
The triple, and the grid that is not this task
awsKey, awsIv, and awsContext are documented as all three or none. The minimal request in the docs sends websiteURL only. Start there. Add the triple only when the challenge page has already given you all three values. A partial set is ERROR_BAD_REQUEST, not a more precise solve.
If the page is a 3 by 3 image grid instead of a token challenge, switch types. AwsWafClassification takes images (nine base64 tiles, in order) and question (aws:grid:bag, or a bare bag). It is async, so that one does poll. The solution is objects, indices 0 through 8, row-major, top-left first. Supported targets are bag, bed, bucket, chair, clock, curtain, and hat. A target outside that list returns ERROR_CAPTCHA_UNSOLVABLE and is not charged.
Cost, and the errors worth handling
The pricing page lists this type at under 3 seconds and $0.60 per 1,000 successful solves, the same rate as the grid type. Failed and timed-out tasks cost nothing. There is no monthly fee, and top-up credit does not expire. New accounts start with $10 in test credit, which is 16,666 successful solves at this rate.
On ERROR_CAPTCHA_UNSOLVABLE, do not retry the same parameters in a loop. The error dictionary says to inspect the site and the fields, then create a new task. ERROR_TASK_NOT_SUPPORTED means the type string is wrong or the type is disabled. ERROR_ZERO_BALANCE means stop and top up. ERROR_BAD_REQUEST means a field is missing or malformed, which is the partial-triple case above.
Related, and not the same call
- Site: kaptra.ai
- AntiAwsWafTask docs: kaptra.ai/docs
- A widget that does poll: How to solve Cloudflare Turnstile in Python
- Cookies plus your own proxy: How to solve the Cloudflare 5-second challenge in Python
- A token with a two-minute clock: How to solve reCAPTCHA v2 in Python
About the author
- Official account of Kaptra, a captcha recognition API for developers. We publish practical guides on Cloudflare Turnstile, Cloudflare Challenge, AWS WAF, and DataDome for test automation and authorised data collection workflows.
Comments
Join the discussionNo comments yet. Be the first to comment.



