mcp-humanizer
This server does not currently have a deployable release on Glama. Follow the repository's setup instructions to install and configure it for your MCP client, including any required credentials and local dependencies.
mcp-humanizer
MCP server providing psycholinguistic, lexical, structural, and discourse-level rules for producing naturally human-sounding written content. Part of the Human, an Education Collective MCP ecosystem.
Demo
Real output from humanizer_check_text on a stock piece of marketing copy. Two hard blocks, seven findings to clear, each with a fix. The server flags and suggests; it never rewrites the text.
Try it live in your browser → The checker is ported to TypeScript and runs entirely client-side, with no server and no network. It is verified to produce the same findings as the Python engine (see web/).
Related MCP server: Natural Voice MCP
What It Does
The reference Claude consults when writing or revising content that needs to read as human-authored. It supplies rules, patterns, substitutions, and before/after examples. It does not rewrite text itself.
Data Dimensions
File | Contents |
| Absolute rules (AR-001/002/003), core principle, two axes, priority hierarchy, application protocol |
| Flagged words, substitutions, vocabulary rules |
| Sentence length, syntax, punctuation, POS targets |
| Neutral bias, emotional layering, subjectivity |
| Transitions, markers, cohesive devices, repetition |
| Cognitive load, self-monitoring, retrieval |
| Per-content-type adjustments |
| Before/after AI → human rewrites |
| ESL, detector brittleness, ethical considerations |
| Human-likeness self-review rubric the calling model runs on its own draft (the |
Tools
This is a public read-only MCP. There are no runtime write tools. Data updates flow through a git push to this repo, then a container restart on the host. There is no tool-level auth.
Reference readers (12): humanizer_get_guide (full composite; about 40k tokens, so it accepts an optional part of 'core', 'lexical', 'anti_patterns' or 'caveats' for clients that cap tool output), humanizer_get_summary (lightweight), humanizer_get_application_protocol (ordered traversal plan), humanizer_get_foundation, humanizer_get_lexical_patterns, humanizer_get_structural_patterns, humanizer_get_sentiment_tone, humanizer_get_discourse_cohesion, humanizer_get_psycholinguistic_texture, humanizer_get_content_profile, humanizer_get_anti_patterns, humanizer_get_caveats
Checker (1): humanizer_check_text runs a deterministic compliance check. Hard violations (block prohibitions_clear): em-dashes, stacked punctuation, banned phrases. Clearable findings (must_clear): comma splices and other punctuation issues, flagged terms, sentence-level banned structures (the BS-* regex/heuristic tells), document-level budget overruns for structures that are ordinary once and a tic three times, rule-of-three density, cross-section uniformity, and reality-assertion (credibility) density. An optional content_type of 'label' or 'notes' tunes the report for headings, worksheet cells and other short-form copy: sentence-length numbers are reported under metrics and are never findings, and segment-uniformity is not reported as a finding, since sentence-length variance is undefined at that length; every other layer still runs. The report's manual_review carries one short line per self-review item for the content type; pass rubric="full" on the last check of the assembled document for each item's full text. Read-only with respect to server state: it analyzes the supplied text and returns a findings report; it does not rewrite or store anything.
Quick Start
pip install -r requirements.lock
PORT=8016 python src/server.pyRunning the checker locally
The rule files load from the repo's data/ automatically when /app/data (the container path) doesn't exist, so a local checkout needs no setup. Set HUMANIZER_DATA_DIR only to point at a different data directory.
PYTHONPATH=. python -c "from src.checker import run_checks; import json; print(json.dumps(run_checks('Your draft here.'), indent=2))"Regenerating the lock file
requirements.txt is the human-edited input with loose top-level bounds. requirements.lock is the deploy artifact and is what the Dockerfile installs. The deployed image runs Python 3.12, so the lock must be resolved against Python 3.12. A lock generated under a different Python version can pin wheels that don't exist for cp312, which breaks the Docker build.
Recommended (uv handles the Python version explicitly):
uv pip compile --python-version 3.12 requirements.txt -o requirements.lockAlternative (pip-tools, which must run inside a Python 3.12 environment):
python3.12 -m venv .venv-3.12
source .venv-3.12/bin/activate
pip install pip-tools
pip-compile --output-file=requirements.lock requirements.txt
deactivateDeployment
See docs/DEPLOYMENT.md for the Docker Compose plus reverse-proxy or tunnel
setup. The server is stateless and read-only, so it also runs fine as a single
local container for development.
Endpoint
Once deployed, connect any MCP client to https://<your-host>/mcp with no auth.
Tool Schema Changelog
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
This server cannot be deployed
Maintenance
Related MCP Connectors
Build a voice from your own texts, measure whether a text still sounds like you, then write in it.
DraftCheck: flags AI-writing patterns in prose with fix hints. Deterministic linter, not a detector.
Prose linter + AI-slop detector: weasel words, passive voice, hedging, and research-cited AI tells
Rewrites AI text to read naturally. One tool: humanize(text, intensity). Free plan: 2,500 words/mo.
Related MCP Servers
- AlicenseCqualityDmaintenanceHelps refine AI-generated content to sound more natural and human-like. Built with advanced AI detection and text enhancement capabilities.164MIT
- AlicenseNot gradedqualityNot gradedmaintenanceProvides tools and resources to detect AI-generated writing patterns and refine text for more authentic, human-like communication. It enables users to analyze phrasing via a scoring system and apply conversational guides tailored for platforms like Twitter and LinkedIn.1-
- AlicenseAqualityCmaintenanceEnables detection and elimination of AI slop in text, providing tools to analyze writing for overused phrases, structural issues, and verbosity, and offers human writing rules tailored to context.33MIT
- AlicenseAqualityDmaintenanceEnables human-like non-fiction authoring by training custom voices, applying tone presets, auditing drafts for AI tells, and managing dictionaries, all locally.19117 npm9MIT