Patent MCP Server
🚀 中国专利最准确的开源 MCP。 Give your AI agent the ability to read CN patents with real accuracy — plus global coverage.
Potencialize seu agente
O comando do próprio publicador. Não é um endpoint do PlugMCP.
Host a copy with our keys and logs.
uvx deeparchi-patent-mcp==1.9.2O Claude Code não tem link de instalação para um comando local. Copie o comando abaixo e execute-o em um terminal.
claude mcp add io.github.deeparchi-ai-patent-mcp-server -- uvx deeparchi-patent-mcp==1.9.2Execute em um terminal.
{ "mcpServers": { "io.github.deeparchi-ai-patent-mcp-server": { "command": "uvx", "args": [ "deeparchi-patent-mcp==1.9.2" ] } }}claude_desktop_config.json
URL remota ou um comando na sua máquina? Como as duas instalações diferem
Rodar no PlugMCP
A new server you create. It is not this listing's official endpoint. Use the publisher install when you want their URL and their credentials.
AbrirAn MCP (Model Context Protocol) server that gives AI agents access to patent data — CN patents with CPC-aware correction, plus US/WO global coverage. Runs locally on your machine. No external API, no subscription. Always MIT.
pip install deeparchi-patent-mcp
Or from source:
git clone https://github.com/deeparchi-ai/patent-mcp-server.git
cd patent-mcp-server
pip install -e .
After pip install, the deeparchi-patent-mcp console script is on your PATH.
Add this to your agent platform's MCP config:
{
"mcpServers": {
"patent-mcp": {
"command": "deeparchi-patent-mcp",
"args": []
}
}
}
Same config as Claude Desktop above.
mcp_servers:
patent-mcp:
command: "deeparchi-patent-mcp"
BigQuery is optional. Without
GCP_PROJECT_IDthe server starts normally and the web-backed tools work with no credentials. BigQuery-backed tools return a message explaining how to enable them. To turn them on, add an env var:{ "mcpServers": { "patent-mcp": { "command": "deeparchi-patent-mcp", "args": [], "env": { "GCP_PROJECT_ID": "your-gcp-project" } } } }
{
"mcpServers": {
"patent-mcp": {
"command": "python",
"args": ["-m", "server"],
"cwd": "/path/to/patent-mcp-server/src"
}
}
}
Now ask your agent:
"Get patent US-7650331-B1 and summarize the claims."
The first two cover 80% of use cases. Zero cost. Zero setup.
Three-layer discovery for Chinese patents:
query="芯片" + country=CN searches both English AND Chinese abstracts (v1.5.2 fix).assignee="BOE" + country=CN → company/city-level patent landscape.H01L) for broader CN coverage; specific CPC codes (H01L25/065) trigger web fallback.SEARCH_GUIDE.md for detailed search strategy and tested CPC codes.If you need search_patents, add a GCP project:
export GOOGLE_APPLICATION_CREDENTIALS="/path/to/key.json"
export GCP_PROJECT_ID="your-project-id"
cp run.sh.example run.sh
# Edit run.sh → set your GCP paths
BigQuery free tier: 1 TB/month — individual use is essentially free.
Need multiple people to share one patent-mcp instance? Start it as an HTTP server:
cp run-http.sh.example run-http.sh
# Edit → set GCP creds (skip if only using web tools)
PORT=8090 ./run-http.sh
Team members connect with:
mcp_servers:
patent-mcp:
url: "http://<server-ip>:8090/sse"
A systemd service template is included for production deployment.
┌──────────────┐ ┌───────────────────────────────────────┐
│ AI Agent │────▶│ patent-mcp-server │
│ (Claude, │ │ (runs on YOUR machine) │
│ Cursor, │ │ │
│ Hermes) │ │ search_patents: │
│ │ │ ┌──────────┐ ┌───────────────┐ │
│ │ │ │ BigQuery │───▶│ Firecrawl │ │
│ │ │ │ (primary)│ │ (CN fallback) │ │
│ │ │ └──────────┘ └───────┬───────┘ │
│ │ │ │ │
│ │ │ get_patent / get_patent_claims: │
│ │ │ ┌──────────────────────┐ │
│ │ │ │ Google Patents (web) │ │
│ │ │ │ → BigQuery fallback │ │
│ │ │ └──────────────────────┘ │
└──────────────┘ └───────────────────────────────────────┘
get_patentget_patent(publication_number="US-7650331-B1")
Returns: classifications, citations (X/Y/A/D prior art markers), family ID, dates, inventors, assignees. Cites prior art markers so your agent can assess novelty at a glance.
CN patent note: Google Patents web scraping provides machine-translated English data for CN patents. CPC codes from web scraping are empty (JS-rendered). For CPC, use BigQuery path.
get_patent_claimsget_patent_claims(publication_number="US-7650331-B1")
Returns: full claims text. Supports US, CN (machine-translated English), and most countries via Google Patents web scraping.
search_patentssearch_patents(cpc="G06N", country="CN", after="2023-01-01", limit=5)
search_patents(assignee="TSMC", country="CN")
search_patents(query="芯片", country="CN") # keyword search (CN: searches both EN+ZH)
Search 1.4B patents by keyword, CPC classification, assignee, country, date range. For CN patents, keyword search scans both English and Chinese abstracts.
Cost control: Queries require at least one filter (cpc/country/assignee/after). A dry-run budget guard rejects queries over 50 GB. When BigQuery rejects a CN CPC query (e.g., H01L25/065 → 256 GB), the web fallback automatically searches via Firecrawl.
See SEARCH_GUIDE.md for best practices and docs/cn-cpc-correction-table.md for tested CPC codes.
pip install -e ".[dev]"
pytest tests/ -v # 32 tests, ~1.5s
ruff check src/ tests/ # lint
mypy src/ # type check
MIT — see LICENSE.
DeepArchi OPC — AI agent infrastructure for enterprise architecture.
Patent search over 1.4B global records via Google Patents on BigQuery, with CN coverage.
{ "mcpServers": { "io.github.deeparchi-ai-patent-mcp-server": { "command": "uvx", "args": [ "deeparchi-patent-mcp==1.9.2" ] } }}Ferramentas
Ainda não há lista de ferramentas. Servidores remotos recebem uma após a primeira verificação de saúde.
Perguntas frequentes
Copie nesta página a configuração stdio para Cursor, Claude Desktop, Claude Code ou VS Code. Ela usa a instalação do publicador, não um proxy do PlugMCP.
Este registro não lista nenhuma variável de ambiente obrigatória.
Patent MCP Server roda localmente via stdio.
Última verificação: 2026-11-09. Patent search over 1.4B global records via Google Patents on BigQuery, with CN coverage.