GoHumanize Open Humanizer

io.github.GoHumanize-aiPublished by GitHub account GoHumanize-aiv0.1.12Updated Oct 25, 2026

Rewrite AI-styled text into natural writing with a fine-tune model.

VerifiedSTDIODesktop onlyMedia & DesignAI & ML

Overview

AI-generated overview

Local MCP server that lets an assistant rewrite AI-styled English text into more natural prose using the open GoHumanize Humanizer model.

What you can do
  • Rewrite a passage of English text into more natural human-sounding prose, best at 50 to 400 words and up to 1,500 words paragraph by paragraph
  • Request several rewrites per call and keep the one that changed most while staying a sensible length
  • Ask what the model is and which endpoint is currently in use
When to use it
Useful when you want an assistant to smooth out stiff, machine-sounding English drafts, or when you want to run a small open humanizing model locally through an OpenAI-compatible endpoint.
Requirements
Runs locally via npx (Node.js). Needs an OpenAI-compatible endpoint: either a self-hosted model such as Ollama with the published GGUF build, or the hosted demo endpoint, which requires OPEN_HUMANIZER_API_KEY. Optional variables: OPEN_HUMANIZER_URL, OPEN_HUMANIZER_MODEL, OPEN_HUMANIZER_SAMPLES. Desktop client only.
Before you install
The hosted fallback endpoint is rate-limited, sleeps when idle, and needs OPEN_HUMANIZER_API_KEY; the first request can take one to two minutes and may time out in clients with a 60-second limit. Text you submit is sent to whichever endpoint is configured, so a hosted endpoint means sending your text to a third party. The model is educational and makes no claim about AI detectors.

Installation

In SourceWeft

  1. Open GoHumanize Open Humanizer in the dashboard and add it to a workspace.
  2. Enable the server for the chats that should use its tools.

Desktop only via STDIO. STDIO servers start a local process, so they need the SourceWeft desktop host.

Other MCP clients

Follow the launch instructions in the repository.

README

GoHumanize Open Humanizer MCP server

[npm] [Model on Hugging Face] [DOI] [License: Apache 2.0]

An MCP (Model Context Protocol) server that lets AI assistants call the GoHumanize Open Humanizer: a small open model (Qwen3-4B fine-tune, Apache-2.0) that rewrites AI-styled English text into more natural human prose.

Tools:

  • humanize_text rewrite a passage (50 to 400 words works best; longer texts are processed paragraph by paragraph, up to 1,500 words).
  • about_open_humanizer what the model is and which endpoint is in use.

The model is educational and makes no claim about AI detectors. It is separate from the production models used by GoHumanize.ai.

Several rewrites, best one returned

On modern prose the model sometimes plays safe and hands the text back almost unchanged (about one try in six). Each call asks the endpoint for five rewrites (generated in parallel, so the wait is the same) and keeps the one that moved furthest from the input while staying a sensible length. Set OPEN_HUMANIZER_SAMPLES=1 for a single request.

Links

ResourceLink
Project page and browser demogohumanize.ai/open-model
GoHumanize (the product this research comes from)gohumanize.ai
Model weights and GGUF buildsgohumanize/gohumanize-open-humanizer
Dataset, 2,883 pairs (CC-BY 4.0)gohumanize/gohumanize-open-humanizer-dataset
Dataset on Kagglegohumanize/gohumanize-open-humanizer-dataset
Code and full pipelineGoHumanize-ai/gohumanize-open-humanizer
Write-up: every step, service and resultdocs/paper.md
Archived release, citable DOI10.5281/zenodo.22843083
Python client and CLIpypi.org/project/gohumanize-open-humanizer
MCP server for AI assistantsnpm · source
Training runs, loss curves and configWeights & Biases

Use

json
{  "mcpServers": {    "gohumanize-open-humanizer": {      "command": "npx",      "args": ["-y", "gohumanize-open-humanizer-mcp"]    }  }}

Which endpoint

The model runs wherever you point the server. Running it yourself needs no key and is the recommended setup:

bash
ollama pull hf.co/gohumanize/gohumanize-open-humanizer:Q4_K_Mexport OPEN_HUMANIZER_URL=http://localhost:11434/v1export OPEN_HUMANIZER_MODEL=hf.co/gohumanize/gohumanize-open-humanizer:Q4_K_M

The hosted endpoint sleeps when idle. The first request after a quiet period waits for a GPU cold start, measured at one to two minutes; afterwards a rewrite takes a second or two. MCP clients apply their own timeout, often 60 seconds, so the first call through a client may fail even with a valid key and succeed on retry. Running the model locally avoids this entirely.

The endpoint the server falls back to is the one behind the browser demo on gohumanize.ai/open-model. It is rate-limited and requires OPEN_HUMANIZER_API_KEY, so it is not open for general use; to try the model without installing anything, use the demo on that page.

Any OpenAI-compatible server works:

json
"env": {  "OPEN_HUMANIZER_URL": "http://localhost:11434/v1",  "OPEN_HUMANIZER_MODEL": "gohumanize/open-humanizer"}

OPEN_HUMANIZER_API_KEY sets a bearer token when the endpoint needs one.

Licence

Apache-2.0.

Source: README.md at commit 5caeb64

Tools

0
SourceWeft has not started this package in its sandbox yet, so its tools are not known.

Version history

1
  1. v0.1.12LatestSep 26, 2026

Score

50% of the applicable checks met

How complete this listing is, computed from the registry entry, the endpoint and public sources. Not a rating of the server and not a statement that it is safe or working today: reachability is a dated check, and no installs or reviews go into the score.

  • Reached in the last 30 days25 pts

    Keep the endpoint answering the MCP handshake; SourceWeft probes it every 30 days.

  • Auth declared10 pts

    Declare the credential headers in the registry entry, or keep the endpoint answering the probe so its auth can be classified.

  • Tools listed15 pts

    Answer tools/list without a credential, or ship a package that starts in a sandbox without one.

  • DescriptionRequired15 pts
  • Logo10 pts
  • Documentation15 pts
  • License10 pts