yargi-cli

CLI tool for Turkish legal databases. Designed for AI agents and programmatic use.

Origin: This project is the CLI counterpart of yargi-mcp, a Python-based MCP server that provides access to Turkish legal databases. While yargi-mcp serves LLM applications via the Model Context Protocol, yargi-cli provides the same capabilities as a standalone command-line tool — JSON output, pipe-friendly, zero authentication.

🌍 Türkçe README

Why?

AI agents (LLM tool-use, autonomous coding agents, RAG pipelines) need a simple, predictable interface to query Turkish court decisions. yargi-cli provides:

• JSON-only output — every command writes structured JSON to stdout
• Pipe-friendly — chain with jq, xargs, or any Unix tool
• Rich --help — parameter descriptions, search operators, output schemas, and examples are embedded in help text so agents can self-discover the API
• No auth, no config — just install and call

Supported Databases

Currently implements the Bedesten module (bedesten.adalet.gov.tr):

Court Type	Flag	Description
YARGITAYKARARI	-c YARGITAYKARARI	Yargıtay (Court of Cassation)
DANISTAYKARAR	-c DANISTAYKARAR	Danıştay (Council of State)
YERELHUKUK	-c YERELHUKUK	Local Civil Courts
ISTINAFHUKUK	-c ISTINAFHUKUK	Civil Courts of Appeals
KYB	-c KYB	Extraordinary Appeals (Kanun Yararına Bozma)

Chamber filtering supports 79 codes covering all Yargıtay/Danıştay divisions. Run yargi bedesten search --help for the full list.

Installation

# Requires Node.js >= 24
npm install -g @saidsrc/yargi

Or run from source:

git clone https://github.com/saidsurucu/yargi-cli.git
cd yargi-cli
npm install
npm run build
node bin/yargi.js bedesten search "test"

Usage

Search decisions

# Basic search (defaults: Yargıtay + Danıştay, page 1)
yargi bedesten search "mülkiyet hakkı"

# Filter by court type and chamber
yargi bedesten search "iş kazası" -c YARGITAYKARARI -b H9

# Date range filter
yargi bedesten search "kamulaştırma" --date-start 2024-01-01 --date-end 2024-12-31

# Multiple court types
yargi bedesten search "idari para cezası" -c DANISTAYKARAR YARGITAYKARARI

# Pagination
yargi bedesten search "tazminat" -p 3

Get full decision text

# Fetch document as Markdown
yargi bedesten doc 1123588300

# Extract just the markdown content
yargi bedesten doc 1123588300 | jq -r '.markdownContent'

Pipe examples

# Get first result's document ID
yargi bedesten search "mülkiyet hakkı" | jq -r '.decisions[0].documentId'

# Search → get first result's full text
yargi bedesten search "mülkiyet hakkı" \
  | jq -r '.decisions[0].documentId' \
  | xargs yargi bedesten doc

# Get all case numbers from a search
yargi bedesten search "iş kazası" -c YARGITAYKARARI | jq '[.decisions[] | .esasNo]'

Search operators

Operator	Example	Effect
Simple	"mülkiyet hakkı"	Finds both words
Exact phrase	"\"mülkiyet hakkı\""	Finds exact phrase
Required term	"+mülkiyet hakkı"	Must contain mülkiyet
Exclude	"mülkiyet -kira"	mülkiyet but not kira
AND	"mülkiyet AND hak"	Both required
OR	"mülkiyet OR tapu"	Either acceptable
NOT	"mülkiyet NOT satış"	mülkiyet but not satış

Wildcards (*, ?), regex, fuzzy search (~), and proximity search are not supported.

Output Schemas

Search output

{
  "decisions": [
    {
      "documentId": "1123588300",
      "itemType": { "name": "YARGITAYKARARI", "description": "Yargıtay Kararı" },
      "birimAdi": "1. Hukuk Dairesi",
      "esasNo": "2023/6459",
      "kararNo": "2024/7158",
      "kararTarihiStr": "26.12.2024",
      "kararTarihi": "2024-12-25T21:00:00.000+00:00"
    }
  ],
  "totalRecords": 1988,
  "requestedPage": 1,
  "pageSize": 10,
  "searchedCourts": ["YARGITAYKARARI"]
}

Document output

{
  "documentId": "1123588300",
  "markdownContent": "**1. Hukuk Dairesi  2023/6459 E. ...**\n\n...",
  "sourceUrl": "https://mevzuat.adalet.gov.tr/ictihat/1123588300",
  "mimeType": "text/html"
}

For AI Agents

This CLI is designed to be called by AI agents as a tool. Key points:

1. Self-documenting: Run yargi bedesten search --help or yargi bedesten doc --help to get full parameter descriptions, valid values, output schemas, and usage examples
2. Predictable output: Always JSON to stdout, errors included as {"error": "..."} with non-zero exit code
3. No interactive prompts: Never asks for input, never writes to stderr for progress
4. Stateless: Each invocation is independent, no sessions or cookies

Typical agent workflow

1. yargi bedesten search "<query>" [-c ...] [-b ...] [--date-start ...] [--date-end ...]
2. Parse JSON → extract documentId from decisions array
3. yargi bedesten doc <documentId>
4. Parse JSON → use markdownContent for analysis

Dependencies

Package	Purpose
commander	CLI framework
turndown	HTML → Markdown conversion

No HTTP libraries — uses Node.js native fetch. No UI libraries — output is raw JSON.

License

MIT