Setup Patent Creator System

Complete one-time setup for all patent creator capabilities.

What This Does

Sets up the complete patent creator system:

Download MPEP PDFs (500MB) - USPTO examination manual
Download Legal Sources - 35 USC, 37 CFR, Federal Register updates
Build Search Index - FAISS + BM25 hybrid search (5-65 min depending on GPU)
Configure BigQuery (optional) - Access to 76M+ patents
Install Dependencies - Graphviz for diagrams
Verify Installation - Run test suite

Total Time: 30-90 minutes (most time is index building)

Requirements

Before running:

Python 3.9-3.12 (3.11 recommended)
8GB+ RAM (16GB for index building)
5GB free disk space
Internet connection (for downloads)
Optional: NVIDIA GPU (10x faster index building)

Process

Step 1: Verify Prerequisites

I'll check:

python --version  # Must be 3.9-3.12
pip --version

If Python not found or wrong version, I'll provide install instructions.

Step 2: Install Python Dependencies

I'll install required packages:

cd ${CLAUDE_PLUGIN_ROOT}
pip install -e .

Dependencies installed:

PyTorch (GPU or CPU version based on hardware)
FAISS (vector search)
Sentence Transformers (embeddings)
Rank-BM25 (keyword search)
Pydantic (validation)
Google Cloud BigQuery (optional)
Graphviz Python bindings

GPU Detection:

Checks for NVIDIA CUDA
Checks for Apple Silicon MPS
Falls back to CPU if neither found
Installs appropriate PyTorch version

Step 3: Download MPEP PDFs

I'll download USPTO examination manual:

cd ${CLAUDE_PLUGIN_ROOT}
python mcp_server/server.py --download-mpep

What downloads:

MPEP ZIP file (500MB)
Extracts to 29 PDF files (mpep-0100 through mpep-2900)
Location: ${CLAUDE_PLUGIN_ROOT}/pdfs\

Time: 5-10 minutes depending on connection

Step 4: Download Additional Legal Sources

I'll download supplementary sources:

python mcp_server/server.py --download-all

Downloads:

35 USC Consolidated Patent Laws PDF
37 CFR Consolidated Patent Rules PDF
Federal Register Subsequent Publications PDF

Time: 2-3 minutes

Step 5: Build Search Index

I'll build the hybrid search index:

python mcp_server/server.py --rebuild-index

What happens:

Load all MPEP PDFs
Extract and chunk text (12,543 chunks)
Generate embeddings (BGE-base-en-v1.5, 768-dim)
Build FAISS vector index
Build BM25 keyword index
Save indexes to disk

Time:

With NVIDIA GPU: 5-15 minutes
With Apple Silicon: 10-25 minutes
CPU only: 35-65 minutes

Index location: ${CLAUDE_PLUGIN_ROOT}/python\index\

Disk space: ~1-2 GB for complete index

Step 6: Configure BigQuery (Optional)

For access to 76M+ patents, I'll help configure Google Cloud:

Prerequisites:

Google account (free)
Google Cloud project (free to create)

Setup steps:

# Install Google Cloud SDK
# (I'll provide OS-specific instructions)

# Authenticate
gcloud auth application-default login

# Set project ID
export GOOGLE_CLOUD_PROJECT=your-project-id

Configuration file: I'll create/update .env file:

GOOGLE_CLOUD_PROJECT=your-project-id
PATENT_LOG_LEVEL=WARNING

Cost: BigQuery is FREE for first 1TB/month (>20,000 patent searches)

Step 7: Install Graphviz

For patent diagram generation:

Windows:

choco install graphviz
pip install graphviz

Linux:

sudo apt install graphviz
pip install graphviz

Mac:

brew install graphviz
pip install graphviz

Step 8: Verify Installation

I'll run the test suite:

cd ${CLAUDE_PLUGIN_ROOT}
python scripts/test_install.py

Tests:

[OK] Python version
[OK] MPEP PDFs present (29 files)
[OK] Search index built
[OK] GPU detected (if available)
[OK] Embeddings model loaded
[OK] BigQuery configured (if applicable)
[OK] Graphviz installed
[OK] All Python dependencies

Setup Status

After setup completes, you'll see:

PATENT CREATOR SYSTEM - SETUP COMPLETE
======================================

Core Components:
  [OK] MPEP Search Index (12,543 chunks indexed)
  [OK] GPU Acceleration (NVIDIA CUDA 12.1)
  [OK] Embedding Model (BGE-base-en-v1.5)
  [OK] Hybrid Search (FAISS + BM25)

Optional Components:
  [OK] BigQuery Patent Search (76M+ patents)
  [OK] Diagram Generation (Graphviz)
  [X] USPTO API (not configured)
  [X] Local Patent Corpus (not downloaded)

System Ready!

Available Commands:
  /create-patent          - Create complete patent application
  /search-prior-art       - Search 76M+ patents for prior art
  /review-claims          - Analyze claims for 112(b) compliance
  /research-mpep          - Search MPEP/USC/CFR

Available Skills:
  - mpep-search           - MPEP/USC/CFR research
  - patent-claims-analyzer - Claims compliance checking
  - bigquery-patent-search - Patent search across 76M+ patents
  - patent-diagram-generator - Technical diagrams
  - prior-art-search      - 7-step search methodology
  - patent-application-creator - Complete patent creation

Available Agents:
  - patent-researcher     - Prior art search specialist
  - patent-drafter        - Claims and specification drafting
  - mpep-expert          - USPTO law and regulations
  - patent-illustrator    - Technical diagram creation

Try it:
  /search-prior-art
  [Describe your invention]

Troubleshooting

MPEP Download Fails

Problem: Download timeout or connection error

Solution 1: Retry with longer timeout

python mcp_server/server.py --download-mpep

Solution 2: Manual download

Visit: https://www.uspto.gov/web/offices/pac/mpep/index.html
Download MPEP PDF files
Place in: ${CLAUDE_PLUGIN_ROOT}/pdfs\

Index Building Fails

Problem: Out of memory during index building

Solution 1: Close other applications Solution 2: Increase system memory Solution 3: Use CPU-only mode (slower but less memory)

PATENT_MPEP_DEVICE=cpu python mcp_server/server.py --rebuild-index

GPU Not Detected

Problem: Have GPU but using CPU

Solution 1: Check CUDA installation

nvidia-smi  # Should show GPU

Solution 2: Reinstall PyTorch with CUDA

pip uninstall torch
pip install torch --index-url https://download.pytorch.org/whl/cu121

BigQuery Not Working

Problem: BigQuery searches fail

Solution 1: Check authentication

gcloud auth application-default login

Solution 2: Verify project ID

echo $GOOGLE_CLOUD_PROJECT  # Should show your project ID

Solution 3: Check API enabled Visit: https://console.cloud.google.com/apis/library/bigquery.googleapis.com

Configuration Options

Create .env file in ${CLAUDE_PLUGIN_ROOT}\:

# Google Cloud (for BigQuery patent search)
GOOGLE_CLOUD_PROJECT=your-project-id

# USPTO API (optional)
PATENT_USPTO_API_KEY=your-api-key-from-data.uspto.gov

# Search Configuration
PATENT_MPEP_USE_HYDE=true        # Use HyDE query expansion
PATENT_MPEP_DEVICE=cuda          # cuda, mps, or cpu
PATENT_LOG_LEVEL=WARNING         # DEBUG, INFO, WARNING, ERROR

# Performance
PATENT_MPEP_BATCH_SIZE=32        # Embedding batch size
PATENT_MPEP_MAX_LENGTH=512       # Max token length

What Gets Installed

Directory Structure:

${CLAUDE_PLUGIN_ROOT}\
├── pdfs/                          # Source documents
│   ├── mpep-0100.pdf
│   ├── mpep-0200.pdf
│   ├── ...
│   ├── mpep-2900.pdf
│   ├── consolidated_laws.pdf      # 35 USC
│   ├── consolidated_rules.pdf     # 37 CFR
│   └── subsequent_publications.pdf # Updates
├── mcp_server/
│   ├── index/                     # Search indexes
│   │   ├── mpep_index.faiss      # Vector index
│   │   └── mpep_metadata.json    # Metadata + BM25
│   ├── server.py                 # Main server
│   ├── mpep_search.py            # Search implementation
│   ├── claims_analyzer.py        # Claims checker
│   ├── bigquery_search.py        # BigQuery integration
│   └── diagram_generator.py      # Diagram creation
└── .env                          # Configuration

After Setup

You're ready to:

Search MPEP: Use /research-mpep or MPEP Search skill
Search Patents: Use /search-prior-art command
Review Claims: Use /review-claims command
Create Patents: Use /create-patent command
Generate Diagrams: Use Patent Illustrator agent

Next Steps

Try it out:

/research-mpep

> Search MPEP for claim definiteness requirements

---

/search-prior-art

> I have a blockchain-based authentication system...

---

/review-claims

> [Paste your claims]

---

/create-patent

> I'll create a complete patent application...

Re-Running Setup

Safe to run multiple times:

Won't re-download if files exist
Won't rebuild index unless forced
Updates configuration if changed

Force rebuild:

python mcp_server/server.py --rebuild-index

Uninstall

To remove:

# Remove downloaded files
rm -rf ${CLAUDE_PLUGIN_ROOT}/pdfs\
rm -rf ${CLAUDE_PLUGIN_ROOT}/python\index\

# Remove package
pip uninstall claude-patent-creator

/setup-patent-system