Patent Research

Python 3.11+ • MIT License • MCP Server • Ruff • 47 tests

CI PyPI

A generic MCP server for patent research, architecture extraction, claims analysis, and pattern synthesis.

Fetches patents from Google Patents, extracts structured sections, analyzes architectural patterns, assesses claims risk, and synthesizes reusable design patterns — all through Model Context Protocol (MCP) tools.

---

Why

Patent research for system architecture is repetitive and error-prone. This tool automates the pipeline:

fetch → extract → analyze → firewall → pattern → export

Each step produces structured, reviewable artifacts that feed into architecture decision records (ADRs), not unstructured notes.

---

Quick Start

# Install with dev dependencies
python -m venv .venv
source .venv/bin/activate
pip install -e ".[dev]"

# Verify
patent-research seeds

---

CLI Usage

Command	Description
patent-research seeds	List seed patents with metadata
patent-research fetch <number> [--pdf]	Download patent HTML + plain text
patent-research sections <number>	Extract structured sections (abstract, claims, description, etc.)
patent-research patterns	List saved architectural patterns
patent-research export	Generate complete research summary

# Complete workflow
patent-research seeds
patent-research fetch US7979296B2
patent-research sections US7979296B2 --save
patent-research patterns
patent-research export

---

MCP Server

Run as a standard MCP server for use with any MCP client (Claude Desktop, Hermes, etc.):

python -m patent_research_mcp.server

Tools (11 total)

Tool	Description
patent_seed_list	List all seed patents with metadata
patent_fetch	Download patent HTML, plain text, and optional PDF
patent_get_sections	Extract structured sections from a fetched patent
architecture_card_save	Save a structured architecture analysis
claims_firewall_save	Save a claims risk assessment
pattern_save	Save a reusable architecture pattern
pattern_list	List all saved patterns
pattern_compare	Compare patterns across patents for shared concepts
research_export_markdown	Generate a complete research summary
module_proposal	Generate a module proposal template
prompt_get	Get available prompt templates (core + plugin)

Example: Hermes config

mcp:
  servers:
    patent-research:
      command: /path/to/.venv/bin/python
      args: [-m, patent_research_mcp.server]
      enabled: true
      env:
        RESEARCH_PLUGIN: /path/to/private-plugin  # optional

---

Data Flow

Google Patents
     │
     ▼
  fetch()        ─── raw/ (HTML + TXT + optional PDF)
     │
     ▼
  get_sections() ─── sections/ (structured JSON)
     │
     ▼
  ArchitectureCard  ─── cards/ (architecture analysis)
  ClaimsFirewall    ─── claims/ (risk assessment)
  PatternCard       ─── patterns/ (reusable patterns)
     │
     ▼
  export()        ─── exports/ (research summary)

Artifacts

Artifact	Schema	Purpose
ArchitectureCard	ArchitectureCard	Structured architecture analysis: problem, components, ontology, patterns
ClaimsFirewall	ClaimsFirewall	Risk assessment: dangerous claims, safe abstractions, design-around
PatternCard	PatternCard	Reusable pattern: entities, events, states, reusable principle

---

Plugin System

The server supports private plugins via the RESEARCH_PLUGIN environment variable. When set, the plugin's patents.json overrides the default seed list. This allows domain-specific patent collections without forking the generic core.

patent-research-mcp/          # public, generic
└── src/patent_research_mcp/
    └── server.py             # checks $RESEARCH_PLUGIN/patents.json

private-plugin/               # private, domain-specific
└── patents.json              # SeedPatent array

Example patents.json:

[
  {
    "publication_number": "US20220237532A1",
    "title": "Digital Twin of Organizational Processes",
    "domain": "enterprise-architecture",
    "why_it_matters": "...",
    "google_patents_url": "https://patents.google.com/patent/US20220237532A1/"
  }
]

---

Schemas

Class	File	Description
ArchitectureCard	schemas.py	Full patent architecture analysis
ClaimsFirewall	schemas.py	Claims liability assessment
PatternCard	schemas.py	Reusable architectural pattern
SeedPatent	schemas.py	Seed patent entry
CompareResult	schemas.py	Cross-pattern comparison result

---

Environment Variables

| Variable | Default | Description | |---|---| | PATENT_RESEARCH_DATA | $CWD/data | Data storage directory | | RESEARCH_PLUGIN | — | Path to private plugin with custom seeds |

---

Docker

docker build -t patent-research-mcp .
docker run -v $(pwd)/data:/data -p 8000:8000 patent-research-mcp

The container includes Playwright with Chromium for Google Patents fetching.

---

Architecture Decisions

Architectural decisions are recorded as ADRs in docs/adr/.

---

Wiki

Extended documentation is available on the GitHub Wiki:

Page	Description
Home	Overview and quick links
Quick Start	Setup in 2 minutes
MCP Tools Reference	All 11 tools
CLI Reference	Command-line usage
Plugin System	How plugins work
Plugin Development	Build your own
Architecture	System design

---

Contributing

See CONTRIBUTING.md for setup, conventions, and PR process.

---

Development

# Install with dev dependencies
make install

# Run tests (unit only)
make test

# Run all tests including e2e
make test-all

# Run tests with coverage report
make test-cov

# Lint and format
make lint
make format

# Type check
make typecheck

# Build the package
make build

Pre-commit hooks (optional):

pip install pre-commit
pre-commit install

Project Structure

patent-research-mcp/
├── src/patent_research_mcp/
│   ├── __init__.py
│   ├── server.py       — MCP server (tools) + CLI
│   ├── schemas.py      — Pydantic models
│   ├── patents.py      — Google Patents fetcher (Playwright)
│   ├── store.py        — JSON file storage
│   ├── normalizer.py   — Text & synonym normalization
│   ├── exporter.py     — Markdown report generation
│   └── seed.py         — Default seed patent data
├── data/               — Patent artifacts (gitignored except examples)
├── tests/              — pytest suite
└── pyproject.toml      — Build config

---

License

MIT — see LICENSE for details.