WriteHuman AI Humanizer Review: Free Limits, MCP, and File Boundaries
WriteHuman publishes clear text limits for its browser, MCP connector, and REST API, but those limits belong to different workflows. This review separates what the pages state from what was not tested.
HumanPen Team
· 21 min read
What does WriteHuman offer?
WriteHuman is a text-first AI humanizer with a small public free tier, larger paid browser limits, and separate MCP and REST API contracts. The checked pages document 250 words and three monthly humanizer requests for Free, but they do not document an editable DOCX round trip or a future Turnitin result.
I opened WriteHuman's current homepage, pricing page, MCP documentation, API page, About page, and the HumanizerBench disclosure page on August 26, 2026. I did not submit text, sign in, authorize an MCP client, buy a plan, call the API, upload a file, or run a detector.
HumanPen publishes this review and sells the document workflow described near the end. Treat both companies' product statements as first-party descriptions. This is not an independent detector comparison.
The useful comparison starts by keeping the browser, MCP connector, and API in separate columns. Their limits are not interchangeable.
How much can the browser humanizer process?
The public homepage showed a text area with a `0/250` meter before sign-in. The current pricing table then attached a monthly request count and per-request word limit to each browser plan.
| Plan | Humanizer requests per month | Maximum words per request | Output variations |
|---|---|---|---|
| Free | 3 | 250 | 1 |
| Basic | 80 | 600 | 2 |
| Pro | 200 | 1,200 | 3 |
| Ultra | Unlimited | 3,000 | 5 |
These are vendor-published plan limits, not completed runs from this review. `Unlimited` appears in the Ultra monthly-request row, while the same row still limits one browser request to 3,000 words. A 9,000-word paper therefore does not fit into one browser input merely because the monthly request count is unlimited.
Keep two numbers beside every plan: how many requests the plan allows and how much one request can hold.
What changes in MCP and the REST API?
WriteHuman's MCP connector and REST API are separate product surfaces. The MCP page documents a hosted HTTP endpoint with OAuth 2.1 and three tools: `humanize_text`, `detect_ai_text`, and `get_account`. The API page documents bearer-key JSON requests.
| Surface | Humanization input | Published request boundary | Result shape |
|---|---|---|---|
| MCP Pro | Text | 200 humanize requests per month; up to 1,200 words per humanize call | Multiple text variations inside the agent conversation |
| MCP Ultra | Text | 1,000 humanize requests per month; up to 3,000 words per humanize call | Multiple text variations inside the agent conversation |
| REST API Standard | JSON text | 125,000 words per month; 2,000 words per call | One text output per request |
| REST API Premium | JSON text | 400,000 to 3,000,000 words per month; 4,000 words per call | Three text outputs, with the page describing automatic scoring and ranking |
The same word can refer to different meters. Pro's 200 browser requests and 200 MCP humanize requests are presented on different pages. API plans count a monthly word balance rather than browser-plan requests. Do not move a number from one surface into another without the vendor explicitly connecting them.
The MCP documentation says an API subscription can also unlock MCP at API usage and that plan usage can fall back to an API word balance. That is a documented billing path, not permission to treat API per-call limits as MCP per-call limits.
Does the Basic plan include MCP access?
WriteHuman's current public pages disagree at this boundary. The upper Basic card on the checked pricing page includes the text `MCP access`. The dedicated MCP page says the connector is included with Pro, Ultra, and API plans, then says Free and Basic users can upgrade to get it.
This article cannot resolve that conflict from public text alone. The pages may be out of sync, or the pricing card may use a broader label than the connector entitlement described in the setup documentation. Anyone choosing Basic specifically for MCP should verify the entitlement in the live checkout or account before paying.
When two first-party pages disagree, record both. Do not silently pick the one that makes the comparison table cleaner.
What is the document-file boundary?
Every checked WriteHuman humanization path was text-oriented. The homepage exposed a textarea and no file input. The MCP schema requires a `text` field. The REST example sends JSON text and returns JSON text.
| Evidence found | What it supports | What it does not support |
|---|---|---|
| Homepage textarea and `0/250` meter | A public paste-in browser workflow | Successful DOCX upload or download |
| MCP `humanize_text` schema | Text passed from an agent to the hosted connector | A document path, editable-file return, or formatting round trip |
| REST `POST /v1/humanize` example | JSON text in and humanized JSON out | A same-format Word or PowerPoint result |
That does not prove WriteHuman lacks every file feature behind every account screen. It establishes a narrower fact: the public browser, MCP, and API contracts checked for this review did not document an editable DOCX or PPTX round trip. If the deliverable is a formatted paper, pasted output still has to be placed back into the file and checked against citations, fields, tables, and styles.
A text result can finish a chat task. It does not automatically finish a document task.
What do the published human scores prove?
WriteHuman's homepage displays a panel of vendor-reported `Human scores` and links to HumanizerBench. This review does not repeat those scores because no shared sample was run and no detector output was reproduced.
HumanizerBench's current About page says it is operated by WriteHuman, which sells one of the products on the leaderboard. The page also says it publishes methodology, raw data, and a scoring script. Those disclosure and reproducibility statements are useful, but operator disclosure has to stay attached whenever the benchmark is cited.
The benchmark is therefore a disclosed vendor-operated source, not independent evidence for this review. More importantly, its text-output detector scores cannot establish whether a real DOCX keeps its citations, cross-references, tables, or styles.
A published detector score and a successful document round trip answer different questions.
When is HumanPen the relevant workflow?
WriteHuman's public contract is a direct fit when the deliverable is text, the input fits the selected plan, and browser, MCP, or API output can stay as text. Its MCP surface is especially explicit about OAuth, tool names, request counts, and per-call word limits.
HumanPen is relevant when the deliverable itself is an editable DOCX or PPTX. Our public workflow accepts document files, lets the user choose the passages allowed to change, and returns the editable file. Those are HumanPen's own product claims, not results from a shared test in this article.
Whichever route you choose, use a disposable sample before a real paper. For a text tool, check meaning and citations after pasting the result back. For a document tool, reopen the returned file and inspect the document objects as well as the visible prose. Eligible HumanPen results can continue lowering AI for free.
Choose by the artifact you must deliver, then test that exact input-to-output path.
Frequently asked questions
Is WriteHuman free? The current pricing table lists a Free plan with three humanizer requests per month, 250 words per request, one output variation, and ten detector checks per month.
What is WriteHuman's free word limit? The checked public editor showed `0/250`, and the pricing table listed 250 words per request for Free. That is a per-request size, not a monthly word pool.
Does WriteHuman accept DOCX files? The public homepage, MCP schema, and REST example checked for this review were text-oriented and did not document an editable DOCX round trip. This is a checked-surface boundary, not a product-wide impossibility claim.
Does WriteHuman have an MCP server? Yes. Its dedicated MCP page publishes a hosted endpoint, OAuth 2.1 authentication, and humanize, detect, and account tools. That page says MCP is included with Pro, Ultra, and API plans.
Does WriteHuman Basic include MCP? The checked pages conflict. The pricing card includes `MCP access`, while the dedicated MCP page says Basic users must upgrade. Verify the live entitlement before buying Basic for this feature.
Can WriteHuman guarantee a Turnitin result? No such conclusion was tested here. WriteHuman's pages publish vendor claims and limits; Turnitin does not disclose a named list of humanizers it detects, and this review ran no detector benchmark.
Official pages checked
These first-party pages were checked on August 26, 2026. Reopen them before buying or integrating because plan limits and product contracts can change.
KEEP READING