WriteStat MCP Server
MCP server for text readability analysis and AI pattern detection. Helps writers identify AI-like patterns and improve readability.
Created by Du'An Lightfoot | @labeveryday
Installation
pip install writestat-mcp
# Optional: ML-based detection (~500MB for torch/transformers)
pip install writestat-mcp[ml]
# Required: NLTK data
python -c "import nltk; nltk.download('punkt_tab')"
Tools
| Tool | Description |
|---|---|
analyze_text |
Readability metrics (Flesch-Kincaid, SMOG, etc.) |
find_hard_sentences |
Complex sentences with explanations |
check_ai_phrases |
Pattern-based AI detection (60+ patterns) |
detect_ai_ml |
ML detection via GPT-2 perplexity (optional) |
batch_analyze |
Process multiple texts in parallel |
compare_texts |
Before/after comparison |
Claude Desktop Setup
Add to ~/Library/Application Support/Claude/claude_desktop_config.json (macOS) or %APPDATA%/Claude/claude_desktop_config.json (Windows):
{
"mcpServers": {
"readability": {
"command": "writestat-mcp"
}
}
}
With Claude Code
# After PyPI publish
# Pattern detection only (lightweight)
claude mcp add writestat-mcp -- uvx writestat-mcp
# Or with ML detection (~500MB download)
claude mcp add writestat-mcp -- uvx "writestat-mcp[ml]"
# From local source
cd /path/to/writestat-mcp
pip install -e .
claude mcp add writestat-mcp -- writestat-mcp
Example Prompts
Full analysis workflow:
I just wrote this blog post. Check the readability, find any difficult sentences, and flag anything that sounds too AI-generated. Then suggest improvements:
Editing pass:
This is my draft and my revised version. Compare them and tell me if the readability improved and if I removed the AI-sounding phrases. {First_draft} vs {second_draft}
Quick AI check:
Does this paragraph have any AI tells? Be specific about which phrases to fix:
Target audience check:
I'm writing for high school students. Is this text at the right reading level? Which sentences are too complex:
AI Detection: What to Expect
This tool uses heuristic pattern matching and zero-shot perplexity scoring—not a fine-tuned classifier.
How It Works
- Pattern detection: Catches stylistic markers (em dashes, filler phrases, buzzwords)
- ML detection: Measures perplexity, vocabulary diversity, burstiness
Accuracy Context
Research shows fine-tuned RoBERTa models achieve ~99% F1 on ChatGPT detection (Guo et al., 2023). Our lightweight approach won't match that. It's designed for:
- Quick pattern screening
- Catching obvious AI tells
- Educational awareness about AI writing patterns
Not suitable for: Academic integrity decisions, high-stakes verification
What the Research Found
The HC3 paper identified key ChatGPT markers we detect:
- Lower perplexity (more predictable) ✓
- Lower vocabulary diversity ✓
- Formal conjunctions ("Furthermore", "It's important to note") ✓
- Organized structure with clear transitions ✓
Score Interpretation
Readability (Flesch-Kincaid Grade)
| Grade | Audience |
|---|---|
| 5- | Elementary |
| 6-8 | Middle school |
| 9-12 | High school |
| 13+ | College |
AI Probability (ML)
| Score | Interpretation |
|---|---|
| 0-30 | Likely human |
| 30-60 | Uncertain |
| 60-100 | Likely AI |
Requirements
- Python 3.10+
- Core: fastmcp, textstat, nltk
- Optional
[ml]: torch, transformers
License
MIT
References
- HC3: How Close is ChatGPT to Human Experts? - Guo et al., 2023
- Model Context Protocol
- FastMCP
Metadata
Release files for writestat-mcp 0.1.0
For a detailed explanation of source distributions (sdists) and built distributions (wheels), please see the package formats documentation.
Source distribution (sdist)
| File | Size | Uploaded | Provenance | Actions |
|---|---|---|---|---|
| writestat_mcp-0.1.0.tar.gz | 39.9 kB | Without provenance | Details for writestat_mcp-0.1.0.tar.gz |
Built distribution (wheel)
| File | Interpreter | ABI | Platform | Actions Reset |
|---|---|---|---|---|
| writestat_mcp-0.1.0-py3-none-any.whl | Python 3 | none | any | Details for writestat_mcp-0.1.0-py3-none-any.whl |
Total release size: 77.9 kB
Release files / writestat_mcp-0.1.0.tar.gz
| Download URL | writestat_mcp-0.1.0.tar.gz |
|---|---|
| Size | 39.9 kB |
| Tags | Source |
|
SHA-256 checksum How to use checksums |
f8d3149cf3f4d1742fd0b3cc16436bd2fc1612572d9fdc638c6b0381b980c4cf
|
|
BLAKE2b-256 checksum How to use checksums |
79895f85f4e9def05d61feaea7f4b92312453ee21973aa93235b2316ca8807d9
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
No |
| Uploaded via |
twine/6.2.0 CPython/3.13.5
|
Release files / writestat_mcp-0.1.0-py3-none-any.whl
| Download URL | writestat_mcp-0.1.0-py3-none-any.whl |
|---|---|
| Size | 38.0 kB |
| Tags | Python 3 |
|
SHA-256 checksum How to use checksums |
c100683650b67c6787cfe2157b7e121fe3f828bf3f2110a79c07951850b06785
|
|
BLAKE2b-256 checksum How to use checksums |
14161ef439bbe0a1e0bfa121447cc2e71d45543e0b3aede7e0becf41045b098e
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
No |
| Uploaded via |
twine/6.2.0 CPython/3.13.5
|
Security for writestat-mcp 0.1.0
Provenance
Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.
No provenance found
None of the files in this release were uploaded with provenance metadata.
A note on package security
PyPI is a repository for distributing packages, not a service for auditing them. While third-party security researchers and our staff do their best to detect and remove malicious packages, PyPI cannot guarantee the quality, security or suitability of any given package.
To learn more about protecting yourself from malicious actors, visit our package security documentation.