Utility Patent Reviewer

Vision

I have a dream... of a world where inventors can transform their ideas into patent-ready applications without navigating complex legal interfaces or paying prohibitive upfront costs. By combining Claude's reasoning capabilities with patent examination standards, I hope to bridge the gap between invention and protection by turning rough notes, drawings, and concepts into structured, examination-ready patent applications through natural conversation rather than specialized legal expertise.

This HEAVY work-in-progress platform will illuminate the patent landscape, helping inventors understand prior art and competitive positioning before they invest time and resources, enabling smarter strategic decisions about what and how to protect. In turn, it will educate through the drafting process, transparently applying USPTO standards (MPEP) to provide constructive feedback that teaches inventors what examiners look for, without crossing into legal advice.

The system will deliver examination-ready quality by producing 80-90% complete applications that are properly structured, internally consistent, and aligned with formal requirements, ready for final attorney review and filing.

My North Star: An inventor should be able to have a conversation about their idea and walk away with either a patent draft ready for attorney review or clear guidance on how to refine their innovation to be more defensible, all without needing to first become a patent expert themselves.

I want to emphasize that this tool will not replace patent attorneys; I want to make their expertise more accessible and enable inventors to arrive at that conversation better prepared and more informed.

Utility Patent Reviewer

Talk to Claude about your invention. Get patent-ready documentation.

This tool gives Claude AI instant access to US Patent Office rules and 9.2+ million patents, so you can prepare patent applications through natural conversation.

Everything runs on your computer. Your ideas stay private. No subscriptions.

Recent Updates

Automatic GPU Setup: install.py detects your GPU and installs correct PyTorch automatically
RTX 5090/5080 Support: Full compute capability detection with CUDA 12.8
Package Compatibility: Pre-tested compatible versions (no more import errors)
One-Command Install: Just run python install.py - all dependencies handled automatically
USPTO API Integration: Search 11M+ patents live without downloading corpus (free API key)
Automated Reviews: Slash commands for complete patent application compliance checks
Patent Diagrams: Generate patent-style technical drawings using Graphviz

Quick Start

IMPORTANT: This tool is designed to work with Claude Code, Anthropic's official CLI for Claude AI. You cannot use this tool with the Claude web interface, API, or other chat tools. You must have Claude Code installed before proceeding.

Requirements

Claude Code - REQUIRED (this is the AI interface that runs the tool)
Python 3.9+ (check with python --version)
Internet connection (for initial setup only)
Disk space:
- Minimum: 2GB (just USPTO rules)
- Full: 25GB (includes 9.2M patents)

Installation

Step 1: Clone the repository

git clone https://github.com/RobThePCGuy/utility-patent-reviewer.git
cd utility-patent-reviewer

Step 2: Create and activate virtual environment

Note: This step is required for running install.py. After installation, Claude Code will automatically manage the virtual environment for you during normal usage.

Windows (PowerShell):

python -m venv venv
venv\Scripts\activate

Linux/macOS:

python3 -m venv venv
source venv/bin/activate

Step 3: Run the installer

python install.py

The installer will:

✓ Detect your system (Windows/Linux/macOS)
✓ Detect your GPU (NVIDIA with CUDA support)
✓ Install compatible package versions automatically
✓ Install PyTorch with CUDA 12.8 if NVIDIA GPU detected
✓ Optionally download USPTO rules (~500MB)
✓ Optionally download 9.2M patents (~13GB)
✓ Register with Claude Code automatically

Compatible versions installed:

PyTorch 2.9.0+cu128 (GPU) or 2.9.0 (CPU)
Transformers 4.44.0
Sentence Transformers 3.1.0
NumPy 1.26.4
FAISS 1.12.0

First-time setup: 5-20 minutes (depending on what you download)

Step 4: Restart Claude Code and you're ready to go!

Using the Virtual Environment

Important: When using Claude Code, you do not need to manually activate the virtual environment. Claude Code handles this automatically.

Manual activation is only required for advanced operations:

Running install.py directly
Manually updating dependencies with pip
Updating documentation
Running troubleshooting scripts directly
Other direct CLI operations outside of Claude Code

Activate (for advanced operations only):

Windows:

venv\Scripts\activate

Linux/macOS:

source venv/bin/activate

Deactivate (when you're done with advanced operations):

deactivate

What You Can Do

Search USPTO Rules

Ask Claude questions about patent requirements:

"What are the requirements for patent claims?"
"Search MPEP for claim indefiniteness standards"
"What are enablement requirements under 35 USC 112(a)?"

Claude searches the MPEP, 35 USC, and 37 CFR and provides exact citations.

Review Your Patent Application

Use slash commands for automated reviews:

/full-review - Complete application review (claims + spec + formalities)
/review-claims - Check claims for definiteness, antecedent basis, compliance
/review-specification - Verify written description, enablement, best mode
/review-formalities - Check abstract, title, drawings, required sections
/create-patent - Create complete application with automatic validation

Find Prior Art

Search millions of patents for similar inventions:

"Search for prior art on blockchain-based voting systems"
"Find patents about neural network training from the last 5 years"
"Search for inventions in CPC class G06F related to data compression"
"Find patents by Google about machine learning before 2020"
"Analyze Apple's patent portfolio in augmented reality"

Three Search Options:

PatentsView Search API: Real-time searches with advanced filtering by inventor, assignee, classification, citations, and more. Perfect for competitive intelligence and targeted searches. Requires free API key. See PatentsView Search API Setup below.
Local Corpus: Download 9.2M patents for fast offline semantic search. Best for privacy and comprehensive analysis.
USPTO API: Search live USPTO database (11M+ patents) without downloading. See USPTO API Setup below.

Generate Patent Diagrams

Create patent-style technical drawings:

"Create a flowchart showing my authentication process"
"Draw a block diagram of a machine learning training system"
"Generate a system architecture diagram with sensors, processor, and display"

What's Included

Feature	Description	Setup Required
MPEP Search	Search USPTO examination rules	Auto (500MB)
Patent Review	Automated USPTO compliance checks	Auto
PatentsView Search	Advanced patent search with filtering & analytics	Optional API key
USPTO API	Search 11M+ patents live (real-time)	Optional API key
Prior Art Database	Search 9.2M+ patents offline	Optional (13GB)
Diagram Generation	Create patent-style drawings	Optional (Graphviz)
GPU Acceleration	5-10x faster indexing	Auto-detected

System Status & Health Checks

Before using the patent reviewer, check system health:

patent-reviewer status

Output shows:

=================================================================
SYSTEM STATUS - Utility Patent Reviewer
=================================================================

MPEP Index............ ✓ Ready (150,234 chunks, 3.2 MB)
Patent Corpus......... ✗ Not Built
                         Fix: patent-reviewer download-patents --build-index
USPTO API............. ✓ Connected (Key: ***...abc123)
Graphviz.............. ✓ Installed (v12.2.1)
GPU................... ✓ NVIDIA RTX 5090 (24GB)
Disk Space............ ✓ 127GB available

Models:
  Embeddings        ✓
  Reranker          ✓
  Hyde              ✓

OVERALL STATUS: 6/7 components ready (86%)
=================================================================

Health Check Features

Pre-flight validation - Catches 95%+ of config issues before they cause failures
Actionable fixes - Every ✗ includes exact command to fix
Startup warnings - Server shows health status on startup
Diagnostic reports - Save to JSON for troubleshooting

Common Fixes

# MPEP Index missing
patent-reviewer setup

# USPTO API not configured
export USPTO_API_KEY="your_key_here"  # Linux/macOS
set USPTO_API_KEY=your_key_here       # Windows

# Patent corpus not built
patent-reviewer download-patents --build-index

# Graphviz not installed
sudo apt install graphviz  # Linux
brew install graphviz      # macOS
winget install graphviz    # Windows

Usage Guide

Best Practices

Skills vs Commands

Skills (.claude/skills/patent-reviewer/) - Activate automatically:

Just mention relevant topics: "patent", "USPTO", "MPEP", "claims"
No explicit invocation needed
Claude recognizes when patent expertise is needed

Commands (.claude/commands/) - Invoke with / prefix:

/full-review - Complete application review
/review-claims - Claims-only analysis
/review-specification [topic] - Spec review with optional focus area
/review-formalities [type] - Formalities check with optional type filter
/create-patent - Create complete application with validation

Tip: Use commands when you want a structured workflow, or just ask naturally to let skills activate automatically.

Effective Reviews

Start with a full review:

/full-review

Then paste your application sections when prompted. This gives you a comprehensive baseline.

Iterative refinement: After addressing issues, run focused commands:

/review-claims          # Verify claim fixes
/review-formalities abstract  # Check abstract word count

Get MPEP guidance: Ask questions naturally and the skill will search MPEP automatically:

"What does MPEP say about claim definiteness?"
"Show me requirements for enablement under 35 USC 112"

Prior Art Searches

For recent patents (last 5 years): Use USPTO API (fast, always current):

"Search USPTO for patents about [your technology] from 2020-2025"

For comprehensive searches: Use local corpus (offline, semantic search):

"Search prior art for similar inventions to [your description]"
"Find patents in CPC class G06F related to [technology]"

Best approach: Use both and compare results.

Working with Results

Always review MPEP citations:

Each finding includes specific MPEP section references
Use these to understand USPTO requirements directly
Ask for clarification: "Explain MPEP 2173.05(b) in plain English"

Prioritize fixes:

Critical - Must fix before filing (missing antecedent basis, claim support issues)
Important - Strongly recommend (scope issues, vague terms)
Minor - Consider improving (style, additional examples)

Request revisions:

"Rewrite claim 1 to fix the indefiniteness issue"
"Suggest better wording for element X"

Performance Tips

GPU acceleration:

Install with install.py for automatic GPU detection
GPU speeds up searches 5-10x
Check status: patent-reviewer status

Index building:

MPEP: 5-10 minutes (do this first)
Patents: ~27 hours on RTX 5090 GPU, ~270 hours on CPU (optional, do overnight)
Both are one-time operations

API rate limits:

USPTO API: ~100 requests/minute
Spread searches out if hitting limits
Local corpus has no rate limits

PatentsView Search API Setup

Advanced patent searching with filtering by inventor, assignee, classification, citations, and more. Perfect for competitive intelligence and targeted prior art searches.

Quick Setup (2 minutes):

Request free API key at PatentsView Help Center
Set PATENTSVIEW_API_KEY environment variable
Start searching with natural language queries

Complete setup guide: docs/PATENTSVIEW_API_SETUP.md

USPTO API Setup

Access USPTO's live patent database (11M+ patents) in real-time without downloading the full corpus.

Quick Setup (5 minutes):

Create account at data.uspto.gov/myodp
Verify with ID.me
Generate API key
Set USPTO_API_KEY environment variable
Verify with patent-reviewer status

Complete setup guide: docs/USPTO_API_SETUP.md

Patent Search Comparison

Feature	PatentsView Search API	USPTO API	Local Corpus
Patents Available	US patents (current)	11M+ (live)	9.2M (1976-present)
Setup Time	2 minutes	5 minutes	~27 hours (RTX 5090 GPU)
Storage Required	None	None	15-20GB
Advanced Filtering	✓ (inventor, assignee, CPC, citations)	Limited	Semantic only
Competitive Intelligence	✓ Best	Basic	Basic
Citation Network	✓ Native support	No	No
Semantic Search	No	No	✓ Best
Privacy	Shared with PatentsView	Shared with USPTO	✓ 100% local
Best For	Targeted search, analysis	Recent patents, lookups	Privacy, semantic search

Recommendation: Use all three for comprehensive patent research.

Common Questions

Q: Do I need to download all 9.2 million patents?

A: No! You have three options:

USPTO API: Search 11M+ patents online (free API key required, 5 min setup)
Local Corpus: Download 9.2M patents for offline semantic search (15GB, ~27 hours indexing on RTX 5090 GPU)
Both: Use API for recent patents and local corpus for privacy and speed

Q: What if I don't have a GPU?

A: No problem. The installer detects your hardware and uses CPU automatically. GPU is optional for speed.

Q: Is this a replacement for a patent attorney?

A: No. This tool helps you prepare applications and understand USPTO rules, but always consult a registered patent attorney before filing.

Q: Where does my data go?

A: Everything runs on your computer. Your patent information never leaves your machine unless you explicitly share it.

Downloaded PDFs are stored in the pdfs/ directory. After indexing, you can delete them to save space (the index will still work).

Q: What if setup fails?

A: See ADVANCED-README.md for detailed troubleshooting, or open an issue on GitHub.

Q: How current is the patent data?

A: PatentsView: Updated quarterly. USPTO API: Real-time.

Q: Is my data private?

A: Yes, completely private when using local corpus. USPTO API queries are visible to USPTO.

Q: Can I search foreign patents?

A: Currently US patents only. EPO/WIPO support is on the roadmap.

Platform-Specific Notes

Windows

Run install.py from PowerShell or Command Prompt
Graphviz: winget install graphviz (for diagrams)
GPU support: Auto-detected for NVIDIA cards

RTX 5090/5080 Users: Automatically detected and configured with CUDA 12.8 during installation.

Linux

Some distros require: sudo apt install python3-venv
Graphviz: sudo apt install graphviz (for diagrams)
GPU support: NVIDIA GPUs with CUDA automatically detected

macOS

Works on both Intel and Apple Silicon
Graphviz: brew install graphviz (for diagrams)
GPU support: CPU only (NVIDIA CUDA not available on macOS)

Troubleshooting

"externally-managed-environment" error (Linux)

sudo apt install python3-venv
python3 -m venv venv
source venv/bin/activate

Virtual environment won't activate (Windows)

Set-ExecutionPolicy -ExecutionPolicy RemoteSigned -Scope CurrentUser
venv\Scripts\activate

Graphviz not found

Install Graphviz for your platform:

Windows: winget install graphviz
macOS: brew install graphviz
Linux: sudo apt install graphviz

See docs/GPU_SETUP.md for detailed instructions.

MCP server not showing in Claude Code

claude mcp list

If not listed, the installer will show the manual registration command.

GPU not detected after installation

The install.py script automatically detects and configures GPU support. To verify:

patent-reviewer status

This will show GPU detection status, CUDA availability, and all component health.

If GPU isn't detected but you have an NVIDIA GPU:

Check NVIDIA drivers are installed:

nvidia-smi

Re-run the installer:

python install.py

The installer will automatically:

Detect your GPU compute capability
Install PyTorch with CUDA 12.8 for RTX 5090/5080
Install compatible package versions
Verify everything works

Manual GPU installation (advanced):

# Remove CPU version
pip uninstall torch torchvision torchaudio

# Install CUDA 12.8 version
pip install torch torchvision --index-url https://download.pytorch.org/whl/cu128

# Verify
patent-reviewer status

See ADVANCED-README.md for advanced GPU configuration.

Skill not activating?

Use explicit trigger words: "patent", "USPTO", "MPEP"
Or invoke commands directly with /

MCP tools not available?

claude mcp list                    # Verify registration
patent-reviewer verify-config      # Check paths
patent-reviewer status --verbose   # Check all dependencies

Empty search results?

Check API key: echo $USPTO_API_KEY (Linux/Mac) or $env:USPTO_API_KEY (Windows)
Verify index built: patent-reviewer status
Try broader search terms

Documentation

Main Documentation

README.md (this file) - Complete user guide with setup, features, and troubleshooting
ADVANCED-README.md - Technical architecture, API reference, and development guide

Specific Topics

Environment Variables - docs/ENVIRONMENT_VARIABLES.md
GPU Setup - docs/GPU_SETUP.md
PatentsView Data Access - docs/PATENTSVIEW_DATA_ACCESS.md
PatentsView Search API - docs/PATENTSVIEW_API_SETUP.md
USPTO API - docs/USPTO_API_SETUP.md
Manual Installation - ADVANCED-README.md
API Reference - ADVANCED-README.md
Development - ADVANCED-README.md

Which Guide Should I Use?

New users: Start here (README.md) for complete setup and usage

Developers/Contributors: See ADVANCED-README.md for:

Architecture details
API documentation
Performance tuning
Contributing guidelines
Manual installation procedures

Privacy and Security

All processing is local except USPTO API calls
Your application text never leaves your computer
API calls only send search queries, not your full application
Patent corpus is stored locally for 100% offline operation

Need Help?

Documentation:
- Quick start: This README
- Technical details: ADVANCED-README.md
GitHub Issues: Report bugs or request features
Discussions: Ask questions

Acknowledgments

This tool is built on the work of many open source contributors and data providers:

Data Sources

USPTO (United States Patent and Trademark Office) - Manual of Patent Examining Procedure (MPEP), 35 USC, 37 CFR, and patent database
PatentsView - Bulk patent datasets (9.2M+ patents, 1976-present)
- Data source: S3 bulk downloads at s3.amazonaws.com/data.patentsview.org/download/
- Format: Tab-separated values (TSV) in ZIP archives
- Files: g_patent.tsv.zip (217 MB), g_patent_abstract.tsv.zip (1.6 GB), and related tables
- Last verified: November 2025 (all download URLs working)

AI Models

BGE-base English v1.5 (BAAI, 2024) - Semantic embedding model for document search
MS MARCO MiniLM v2 Cross-Encoder (Microsoft) - Neural reranking model for search relevance

Supporting Libraries

PyTorch - Deep learning framework
Sentence Transformers - Embedding model infrastructure
FAISS - Vector similarity search (Meta AI)
Graphviz - Patent diagram generation

Thank you to all the researchers, developers, and organizations who make open source AI and data access possible.

License

MIT License - See LICENSE for details.

Disclaimer

This tool provides access to USPTO rules and patent databases to assist in patent application preparation. It does not constitute legal advice. Patent applications are complex legal documents. Always consult with a registered patent attorney or agent before filing with the USPTO.

This project is not affiliated with, endorsed by, or sponsored by the United States Patent and Trademark Office (USPTO).

The authors make no warranties about the accuracy, completeness, or suitability of the information provided. Use at your own risk.

Star on GitHub if this helps you! github.com/RobThePCGuy/utility-patent-reviewer

Name		Name	Last commit message	Last commit date
Latest commit History 22 Commits
.claude		.claude
.github		.github
docs		docs
mcp_server		mcp_server
scripts		scripts
.gitattributes		.gitattributes
.gitignore		.gitignore
.mcp.json.template		.mcp.json.template
.typos.toml		.typos.toml
ADVANCED-README.md		ADVANCED-README.md
CHANGELOG.md		CHANGELOG.md
CONTRIBUTING.md		CONTRIBUTING.md
LICENSE		LICENSE
README.md		README.md
SECURITY.md		SECURITY.md
install-gpu.bat		install-gpu.bat
install-gpu.sh		install-gpu.sh
install.py		install.py
package.json		package.json
pyproject.toml		pyproject.toml
requirements.txt		requirements.txt

License

RobThePCGuy/utility-patent-reviewer

Folders and files

Latest commit

History

Repository files navigation

Vision

Utility Patent Reviewer

Recent Updates

Quick Start

Requirements

Installation

Using the Virtual Environment

What You Can Do

Search USPTO Rules

Review Your Patent Application

Find Prior Art

Generate Patent Diagrams

What's Included

System Status & Health Checks

Health Check Features

Common Fixes

Usage Guide

Best Practices

Skills vs Commands

Effective Reviews

Prior Art Searches

Working with Results

Performance Tips

PatentsView Search API Setup

USPTO API Setup

Patent Search Comparison

Common Questions

Platform-Specific Notes

Windows

Linux

macOS

Troubleshooting

"externally-managed-environment" error (Linux)

Virtual environment won't activate (Windows)

Graphviz not found

MCP server not showing in Claude Code

GPU not detected after installation

Skill not activating?

MCP tools not available?

Empty search results?

Documentation

Main Documentation

Specific Topics

Which Guide Should I Use?

Privacy and Security

Need Help?

Acknowledgments

Data Sources

AI Models

Supporting Libraries

License

Disclaimer

About

Topics

Resources

License

Contributing

Security policy

Uh oh!

Stars

Watchers

Forks

Uh oh!

Contributors 5

Uh oh!

Languages