MCP Taiwan Judgment Search
Enables searching and retrieving Taiwan judicial judgments, including full-text search, document details, PDF download, and legal term lookup via MCP tools.
README
MCP Taiwan Judgment Search
An MCP server for searching Taiwan judicial judgments, exposing AI-callable tools over Model Context Protocol (MCP).
Features
- Four MCP tools — full-text search, full-document fetch by ID, PDF download, and legal-term lookup
- stdio JSON-RPC 2.0 — standard MCP transport protocol
@mcp.tool()decorator — Pydantic-typed tool registration- Two-step scraping — handles the judicial site's iframe-based result rendering
- No-auth public endpoint — 司法院裁判書系統 is fully public; no API keys required
- Pure HTML parser layer —
parser/is decoupled from HTTP and tested offline with saved fixtures
Available Tools
| Tool | Description |
|---|---|
search_judgments |
Full-text keyword search across all judicial judgments. Paginated 20 per page. Returns judgment_id, title, ruling date, case reason, URL, and a text preview per entry. |
get_judgment |
Fetch the complete text and metadata of a single judgment by its judgment_id. Returns both a flat content string and a structured paragraphs list (each entry has id/section/level/heading/text — e.g. 理由.一.(三).2) for precise citation. |
lookup_legal_term |
Look up a legal term in the 司法院裁判書用語辭典. Returns definitions for each applicable legal domain (民事、刑事、行政、家事). Optionally filter by domain. |
get_judgment_pdf |
Return or download a judgment's PDF. When save_to (arg) or MCP_TW_JUDGMENT_DOWNLOAD_DIR (env) is set, the file is saved locally and the path is returned; otherwise only the URL is returned. |
Requirements
- Python
3.12+ - One of:
uv(recommended) orpip
Installation
Option 1 — uvx (no install, runs on demand)
uvx mcp-tw-judgment
Option 2 — pip / uv pip
pip install mcp-tw-judgment
# or
uv pip install mcp-tw-judgment
After install, the mcp-tw-judgment console script is available.
Option 3 — From source
git clone https://github.com/asgard-ai-platform/mcp-tw-judgment.git
cd mcp-tw-judgment
uv sync
uv run mcp-tw-judgment
Usage
The server speaks MCP over stdio. Add it to your client of choice:
Claude Desktop
~/Library/Application Support/Claude/claude_desktop_config.json (macOS)
or %APPDATA%\Claude\claude_desktop_config.json (Windows):
{
"mcpServers": {
"tw-judgment": {
"command": "uvx",
"args": ["mcp-tw-judgment"]
}
}
}
Claude Code
Add to your project's .mcp.json:
{
"mcpServers": {
"tw-judgment": {
"command": "uvx",
"args": ["mcp-tw-judgment"]
}
}
}
Cursor
~/.cursor/mcp.json:
{
"mcpServers": {
"tw-judgment": {
"command": "uvx",
"args": ["mcp-tw-judgment"]
}
}
}
Environment Variables
| Variable | Effect |
|---|---|
MCP_TW_JUDGMENT_DOWNLOAD_DIR |
Default directory for get_judgment_pdf downloads. ~ is expanded. If unset and the tool's save_to arg is None, get_judgment_pdf returns the URL without downloading. |
MCP servers are launched by the MCP client (Claude Desktop, Claude Code, Cursor), not by your shell — exporting a variable in ~/.zshrc or ~/.bashrc will not reach the server. Set variables in the client config's env block:
{
"mcpServers": {
"tw-judgment": {
"command": "uvx",
"args": ["mcp-tw-judgment"],
"env": {
"MCP_TW_JUDGMENT_DOWNLOAD_DIR": "~/Downloads/tw-judgments"
}
}
}
}
Apply this to whichever client config you use (claude_desktop_config.json, .mcp.json, or ~/.cursor/mcp.json). Restart the client after editing.
Example Tool Usage
You: 最近有哪些和著作權有關的判決
AI call: tw-judgment - search_judgments (MCP)(keyword: "著作權")
{
"keyword": "著作權",
"total": 55265,
...
}
Result: 以下是最近的著作權相關判決(共 55,265 筆,以下列出最新 10 件核心案件):
115.04.15 │ 智慧財產及商業法院 │ 114年民著訴52 │ 侵害著作權有關人格權爭議
...
You: 請告訴我 114年民著訴52 的詳細資訊
AI call: tw-judgment - get_judgment (MCP)(judgment_id: "IPCV,114,民著訴,52,20260415,1")
{
"title": "智慧財產及商業法院 114 年度民著訴字第 52 號民事判決",
"date": "民國 115 年 04 月 15 日",
"case_reason": "侵害著作權有關人格權爭議",
"content": "……全文字串……",
"paragraphs": [
{"id": "主文", "section": "主文", "level": 1, "heading": null, "text": "……"},
{"id": "事實及理由.一", "section": "事實及理由", "level": 2, "heading": "原告主張:", "text": ""},
{"id": "事實及理由.一.(一)", "section": "事實及理由", "level": 3, "heading": null, "text": "……"},
...
]
}
Result: 本件爭點為…(AI 會直接引用 事實及理由.一.(一) 這段)
You: 把這份判決的 PDF 抓下來放桌面,我要附在書狀證物裡
AI call: tw-judgment - get_judgment_pdf (MCP)(judgment_id: "IPCV,114,民著訴,52,20260415,1", save_to: "~/Desktop")
{
"judgment_id": "IPCV,114,民著訴,52,20260415,1",
"url": "https://judgment.judicial.gov.tw/FILES/IPCV/114%2c%e6%b0%91%e8%91%97%e8%a8%b4%2c52%2c20260415%2c1.pdf",
"path": "/Users/you/Desktop/IPCV,114,民著訴,52,20260415,1.pdf",
"size_bytes": 245678,
"cached": false
}
Result: 已下載到 ~/Desktop/IPCV,114,民著訴,52,20260415,1.pdf。
Tip: Set
MCP_TW_JUDGMENT_DOWNLOAD_DIR=~/Downloads/tw-judgmentsto have every call download to that folder by default; omitsave_toin the call and the tool returns the URL only (no download).
Project Structure
mcp-tw-judgment/
├── app.py # FastMCP singleton
├── mcp_server.py # Entry point (stdio transport)
├── config/settings.py # API base URL, endpoints, request headers
├── connectors/rest_client.py # HTTP GET helper with retry + encoding detection
├── auth/none.py # No-op auth module (public endpoint)
├── parser/
│ ├── judgment_parser.py # Pure HTML parsers for judgments (no HTTP)
│ └── terms_parser.py # Pure HTML parsers for 用語辭典 (no HTTP)
├── tools/judgment_tools.py # MCP tool definitions
├── tests/
│ ├── fixtures/ # Saved HTML responses for offline unit tests
│ ├── test_judgment_parser.py # Unit tests (no network)
│ ├── test_terms_parser.py # Unit tests for terms parser (no network)
│ └── test_all_tools.py # Tool tests (live API, opt-in via RUN_LIVE_TESTS=1)
└── scripts/auth/test_connection.py
Development
# Setup
uv sync
# Connection check
uv run python scripts/auth/test_connection.py
# Run server locally
uv run mcp-tw-judgment
# Offline tests (parser + tool registration)
uv run python -m unittest tests.test_judgment_parser tests.test_all_tools -v
# Live API tests (hits 司法院 endpoint)
RUN_LIVE_TESTS=1 uv run python -m unittest tests.test_all_tools -v
See CONTRIBUTING.md for adding new tools.
License
MIT License — see LICENSE for details.
Data Source & Disclaimer
This project directly scrapes the 司法院裁判書系統 public search interface — this is not an official API.
Please note: This tool is intended for personal research and ad-hoc queries only. Do not use it for bulk automated access or scraping, as this may place undue load on the judicial system's servers. Use at your own discretion and in accordance with the website's terms of use.
Recommended Servers
playwright-mcp
A Model Context Protocol server that enables LLMs to interact with web pages through structured accessibility snapshots without requiring vision models or screenshots.
Magic Component Platform (MCP)
An AI-powered tool that generates modern UI components from natural language descriptions, integrating with popular IDEs to streamline UI development workflow.
Audiense Insights MCP Server
Enables interaction with Audiense Insights accounts via the Model Context Protocol, facilitating the extraction and analysis of marketing insights and audience data including demographics, behavior, and influencer engagement.
VeyraX MCP
Single MCP tool to connect all your favorite tools: Gmail, Calendar and 40 more.
graphlit-mcp-server
The Model Context Protocol (MCP) Server enables integration between MCP clients and the Graphlit service. Ingest anything from Slack to Gmail to podcast feeds, in addition to web crawling, into a Graphlit project - and then retrieve relevant contents from the MCP client.
Kagi MCP Server
An MCP server that integrates Kagi search capabilities with Claude AI, enabling Claude to perform real-time web searches when answering questions that require up-to-date information.
E2B
Using MCP to run code via e2b.
Neon Database
MCP server for interacting with Neon Management API and databases
Qdrant Server
This repository is an example of how to create a MCP server for Qdrant, a vector search engine.
Exa Search
A Model Context Protocol (MCP) server lets AI assistants like Claude use the Exa AI Search API for web searches. This setup allows AI models to get real-time web information in a safe and controlled way.