API documentation
One key and one credit balance for AI detection and humanizing, over REST or MCP. Base URL: https://aicheckerhumanizer.com/api/v1
Authentication
Create a key on your account page. It starts with ach_live_ and is shown only once. Send it as a Bearer token. Your first key gets 5,000 trial credits for 30 days if your email is verified and isn't a disposable address; each address can claim the trial once.
curl https://aicheckerhumanizer.com/api/v1/balance -H "Authorization: Bearer $AICH_API_KEY"Keys belong to your account: any active key can see the account's tasks and balance. Revoking a key stops it immediately; tasks it already started still finish.
Detect
POST /detect with {"text": "..."}. Up to 10,000 words per request; the answer comes back in the same response. When there's a verdict, the response includes a 0–100 score; branch on verdict.basis: for dominant_language the label is in verdict.label; for per_language each entry of verdict.parts (one or more) has its own label. Sentences come back with their positions; in English and Portuguese they can also have scores and colors, while Spanish and French get a whole-text verdict only and their sentences have score: null. We add up the words in paragraphs detected as a supported language. Scoring needs at least the lowest minimum among the detected languages (150 words if a paragraph is detected as English, 200 otherwise); below that, sentences is empty; once scoring runs, sentences can have entries even when no language has enough scored words for a verdict.
Cost: 1 credit per scored word (scoredWords), only when there's a verdict. When verdict is null, reason says why: too_short (a verdict needs at least 150 words in English, or 200 in Spanish, Portuguese and French), no_language_long_enough (several languages, none long enough on its own) or too_many_unscored (worth retrying). None of these are charged.
# essay.json: {"text": "..."} with at least 150 words of English
curl https://aicheckerhumanizer.com/api/v1/detect \
-H "Authorization: Bearer $AICH_API_KEY" \
-H "Content-Type: application/json" \
-H "Idempotency-Key: essay-1182" \
-d @essay.jsonimport os, requests
response = requests.post(
"https://aicheckerhumanizer.com/api/v1/detect",
headers={"Authorization": f"Bearer {os.environ['AICH_API_KEY']}"},
json={"text": text},
timeout=70,
)
result = response.json()
print(result["score"], result["verdict"])const response = await fetch("https://aicheckerhumanizer.com/api/v1/detect", {
method: "POST",
headers: {
Authorization: `Bearer ${process.env.AICH_API_KEY}`,
"Content-Type": "application/json",
},
body: JSON.stringify({ text }),
});
const result = await response.json();{
"score": 100,
"verdict": { "basis": "dominant_language", "language": "en", "words": 207, "label": "likely_ai" },
"reason": null,
"supportedWords": 207,
"scoredWords": 207,
"languages": [
{ "language": "en", "sentenceLevel": true, "supportedWords": 207, "scoredWords": 207, "score": 100, "label": "likely_ai" }
],
"sentences": [
{ "start": 0, "end": 95, "text": "Remote work has fundamentally transformed the way modern organizations operate and communicate.", "language": "en", "words": 12, "score": 100, "color": "ai" }
],
"creditsCharged": 207,
"creditsRemaining": 249793
}| Field | Meaning |
|---|---|
| score | 0–100 for the whole text; higher means more likely AI. Null without a verdict. |
| verdict.label | likely_ai, mixed or likely_human. With several long-enough languages, basis is per_language and each part has its own label. |
| sentences[].start, end | Position in the text you sent, in UTF-16 code units (JavaScript string indexes), end exclusive. text is included so you don't have to slice. |
| sentences[].color | ai, suspicious, human, unscored or unsupported. Null for whole-text-only languages. |
| creditsCharged | Credits charged for the original request. A replay uses no additional credits. |
Humanize
POST /humanize with text, an optional mode (standard, academic, casual, concise; default standard) and strength (light, medium, heavy; default medium). Up to 3,000 words and 200 chunks. It returns 202 with a taskId.
We rewrite eligible paragraphs and split long ones into chunks. You pay only for chunks we successfully rewrite. Quotes, citations, URLs and numbers we recognize are kept as they were; check names in the result, since some can be missed. Paragraphs in a language we don't rewrite, or where the mode isn't available, are left as they were and counted in skippedWords.
Cost: 5 credits per word of each unit that was rewritten, with every unit counted as at least 10 words. Credits are reserved when the task starts (creditsReserved); when it ends you pay only for units with status succeeded, and the rest goes back.
# post.json: {"text": "...", "mode": "casual", "strength": "medium"}
curl https://aicheckerhumanizer.com/api/v1/humanize \
-H "Authorization: Bearer $AICH_API_KEY" \
-H "Content-Type: application/json" \
-H "Idempotency-Key: post-2481" \
-d @post.jsonimport os, time, requests
headers = {"Authorization": f"Bearer {os.environ['AICH_API_KEY']}"}
task = requests.post("https://aicheckerhumanizer.com/api/v1/humanize", headers=headers,
json={"text": text, "mode": "standard"}).json()
while task["status"] in ("queued", "running"):
time.sleep(3)
task = requests.get(f"https://aicheckerhumanizer.com/api/v1/humanize/{task['taskId']}", headers=headers).json()
print(task["output"]){
"taskId": "0f92ace1-57c9-4dd1-a675-5c3967295244",
"status": "queued",
"creditsReserved": 215,
"creditsRemaining": 254785,
"expiresAt": "2026-11-18T19:02:41Z",
"skippedWords": 0,
"output": null,
"units": [
{ "start": 0, "end": 328, "text": "Coordinating stand-up meetings across multiple time zones presents a significant challenge for distributed teams. Our engineers in Berlin, Austin and Manila must navigate overlapping schedules while maintaining productivity. It is crucial to find a meeting time that accommodates everyone without compromising work-life balance.", "status": "pending" }
]
}Get a humanize result
GET /humanize/{taskId}. Poll every 3 seconds or so. A task that hasn't started within 2 minutes ends as failed with reason queue_timeout and costs nothing.
{
"taskId": "0f92ace1-57c9-4dd1-a675-5c3967295244",
"status": "completed",
"output": "Running stand-up meetings across time zones is tough for our distributed team. Engineers in Berlin, Austin, and Manila have to juggle overlapping hours while staying productive—and we need a time that works for everyone without messing with work-life balance.",
"units": [
{ "start": 0, "end": 328, "text": "Coordinating stand-up meetings across multiple time zones presents a significant challenge for distributed teams. Our engineers in Berlin, Austin and Manila must navigate overlapping schedules while maintaining productivity. It is crucial to find a meeting time that accommodates everyone without compromising work-life balance.", "status": "succeeded", "output": "Running stand-up meetings across time zones is tough for our distributed team. Engineers in Berlin, Austin, and Manila have to juggle overlapping hours while staying productive—and we need a time that works for everyone without messing with work-life balance." }
],
"creditsReserved": 215,
"creditsCharged": 215,
"creditsRemaining": 254785,
"expiresAt": "2026-11-18T19:02:41Z"
}| Status | Meaning | Charged |
|---|---|---|
| queued, running | Still working. | Not yet |
| completed | At least one unit was rewritten; all others were left unchanged. | Rewritten units |
| partial | Some units were rewritten, some failed. | Rewritten units |
| unchanged | The text was left unchanged. | 0 |
| failed | Nothing was rewritten. reason is queue_timeout, unavailable or cancelled. | 0 |
| cancelled | A refund or chargeback stopped the task before it started. | 0 |
output is the full text with the rewritten chunks swapped in. The result can be fetched until expiresAt, about 70 minutes after you submit. After that, GET /humanize/{taskId} returns 410 with the task status and any recorded charge while the usage record is kept, about 90 days from submission, and then 404. We delete the encrypted text during cleanup once the result has expired and the task has been finished for at least 10 minutes.
Balance
GET /balance lists credits left and when each batch expires. Credits that expire first are used first.
{
"creditsRemaining": 1245000,
"batches": [
{ "kind": "trial", "credits": 5000, "remaining": 0, "expiresAt": "2026-12-18T10:00:00Z" },
{ "kind": "paid", "credits": 1500000, "remaining": 1245000, "expiresAt": "2027-11-18T10:00:00Z" }
]
}Errors
Errors have an error code you can switch on and a message for people, plus extra fields where they help. 429 and 503 also send a Retry-After header.
{
"error": "insufficient_credits",
"message": "Not enough API credits for this request. Buy more on your account page.",
"creditsRemaining": 1200,
"creditsNeeded": 4100
}| Status | error | When |
|---|---|---|
| 400 | bad_request | The body isn't valid JSON or a field is wrong. |
| 401 | unauthorized | The key is missing, malformed or revoked. |
| 402 | insufficient_credits | Not enough credits. Includes creditsRemaining and creditsNeeded. |
| 402 | trial_paused | Trial credits are paused for the rest of the UTC day. Paid credits still work. |
| 404 | not_found | No task with this ID on your account. |
| 409 | in_progress | The first request with this Idempotency-Key is still running. Retry after retryAfter seconds. |
| 409 | busy | Your account reached its limit of humanize tasks running at once: 10 once you've bought credits, 1 with trial credits only. Wait for one to finish. |
| 410 | result_expired | The result has expired. Status and charges are still included. |
| 413 | too_long | The text is over the word or character limit. Includes maxWords. |
| 413 | too_many_units | The text splits into more than 200 chunks. Includes units and maxUnits. |
| 422 | language_unsupported | None of the text is in a supported language. |
| 422 | mode_unavailable | The mode isn't available for the language of this text. |
| 422 | nothing_to_rewrite | Try a short paragraph in a supported language. A chunk needs at least three words in total and some text we can edit. We skip it if it starts or ends inside protected text such as a quote, citation, URL or number. |
| 422 | idempotency_conflict | This Idempotency-Key was used with a different request. |
| 429 | rate_limited | Hourly limit reached, or too many detections at once. Includes retryAfter. |
| 429 | api_paused | Your account hit its daily limit for work we didn't charge for (failed, unchanged or no-verdict requests). New checks and rewrites pause until midnight UTC; you can still fetch results, check your balance and replay accepted requests. Includes retryAfter. |
| 503 | capacity / detection_unavailable / unavailable | Temporary. Nothing is charged; retry after retryAfter seconds. |
Limits
- Per hour (UTC): 2,000 detections, 600 humanize requests, 20,000 result and balance checks.
- At once: before your first credit purchase, 2 detections and 1 rewrite in progress; after it, 10 detections and 10 rewrites. Queued rewrites count toward the limit.
- Up to 5 active keys per account.
Idempotency
Send an Idempotency-Key header (up to 128 printable characters) with /detect or /humanize. Within 24 hours, reuse the same key, text and options: a finished detection returns its saved scores and your current balance, one still running returns 409, and a failed one can run again; an accepted rewrite keeps its original task ID, and its text expires about 70 minutes after submission. The same key with different content returns 422 idempotency_conflict.
A detection result is replayed even when it had no verdict; use a new key to check the text again. A humanize key keeps pointing to its task whatever the task's outcome; use a new key to start over. If the first request failed before it was accepted, the same key starts fresh. Replays count toward the hourly result-check limit.
MCP server
https://aicheckerhumanizer.com/api/mcp speaks MCP over Streamable HTTP and takes the same key. Tools: detect_ai_text, humanize_text, get_humanize_result and get_balance. Credits, limits and idempotency are the same as REST; pass idempotency_key as a tool argument.
For a new task, humanize_text waits up to 50 seconds and returns the rewrite if it's done. If it's still running, pass the returned taskId as the task_id argument to get_humanize_result. If the client times out or disconnects, the task keeps running and is charged as usual.
claude mcp add --transport http aicheckerhumanizer \
https://aicheckerhumanizer.com/api/mcp \
--header "Authorization: Bearer ach_live_..."{
"mcpServers": {
"aicheckerhumanizer": {
"url": "https://aicheckerhumanizer.com/api/mcp",
"headers": { "Authorization": "Bearer ach_live_..." }
}
}
}OpenAPI
The full description is at https://aicheckerhumanizer.com/api/v1/openapi.json (OpenAPI 3.1); import it into Postman or an SDK generator. Prices are on the pricing page. For an overview, see the AI humanizer API and the AI detector API.