indian-kanoon-mcp 0.1.0

Model Context Protocol (MCP) server for Indian Kanoon

Indian Kanoon MCP Server

An official Model Context Protocol (MCP) server for the Indian Kanoon API, built using the Python FastMCP framework.

This server allows LLMs (like Claude, Cursor, ChatGPT, etc.) to query the Indian Kanoon legal database directly, search for court judgments, download document texts, view citations, and retrieve document fragments.

---

Prerequisites & Installation

The server is built to run seamlessly with uv, Astral's fast Python package installer and runner.

1. Install uv

If you do not have uv installed, you can install it using one of the following methods:

macOS (Homebrew):

brew install uv

Linux/macOS (curl):

curl -LsSf https://astral.sh/uv/install.sh | sh

Windows:

powershell -c "irm https://astral.sh/uv/install.ps1 | iex"

2. Install Project Dependencies

Navigate to the mcp_server directory and synchronize the dependencies:

uv sync

3. Set your API Token

Ensure the INDIANKANOON_API_TOKEN environment variable is set. For local testing, you can export it in your terminal:

export INDIANKANOON_API_TOKEN=your_actual_api_token_here

(Alternatively, you can also place it in a .env file in the directory where you run the server).

---

How to Run & Test the Server

Because this is a standard input/output (stdio) based MCP server, running it directly as a script (e.g. uv run server.py) will wait for JSON-RPC messages and throw a validation error if run interactively in a normal shell. Use the following methods to test and run it correctly:

1. Run the Integration Tests

We have provided a test script to verify that your configuration and API token are working correctly:

uv run python test.py

This runs a search query and fetches document fragments directly from the API.

2. Start the Interactive MCP Inspector

FastMCP comes with a developer console/UI that lets you inspect and trigger the server's tools in your web browser:

uv run mcp dev server.py

This will start the server and print a local URL (e.g., http://localhost:5173 or http://127.0.0.1:6274). Open that URL in your browser to view the interactive console.

Alternatively, you can manually launch the official MCP Inspector using npx:

npx @modelcontextprotocol/inspector uv run server.py

---## Connecting the Server to MCP Clients

You can connect the server to clients without using absolute paths by installing it locally as a tool, or using uvx directly (once the package is published to PyPI).

Option A: Local Tool Installation (Recommended for Local Dev)

Navigate to the mcp_server directory and install the package globally using uv:

uv tool install .

This registers indian-kanoon-mcp as a global executable on your system. You can then run it from anywhere.

Claude Desktop Configuration:

Add this to your claude_desktop_config.json:

{
  "mcpServers": {
    "indian-kanoon": {
      "command": "indian-kanoon-mcp",
      "env": {
        "INDIANKANOON_API_TOKEN": "your_actual_api_token_here"
      }
    }
  }
}

Cursor IDE:

In Cursor Settings -> Features -> MCP:

• Name: indian-kanoon
• Type: stdio
• Command: indian-kanoon-mcp

---

Option B: Using uvx / pip (Once Published to PyPI)

If you publish the package to PyPI (as indian-kanoon-mcp), users won't even need to download this repository. They can run it directly:

Claude Desktop Configuration:

{
  "mcpServers": {
    "indian-kanoon": {
      "command": "uvx",
      "args": [
        "indian-kanoon-mcp"
      ],
      "env": {
        "INDIANKANOON_API_TOKEN": "your_actual_api_token_here"
      }
    }
  }
}

Cursor IDE:

• Name: indian-kanoon
• Type: stdio
• Command: uvx indian-kanoon-mcp

(Alternatively, users can install it globally via pip install indian-kanoon-mcp or uv tool install indian-kanoon-mcp and use indian-kanoon-mcp as the command).

---

Exposed Tools

The server registers four main tools with the LLM:

1. search_kanoon

Search the Indian Kanoon legal database for documents.

• Arguments:
  • query (string, required): The search query. Supports logical operators (ANDD, ORR, NOTT), phrase queries (in quotation marks), and metadata prefixes:
    • doctypes:supremecourt (or delhi, bombay, etc.)
    • fromdate:DD-MM-YYYY / todate:DD-MM-YYYY
    • title:kesavananda
    • cite:citation
    • author:arijit pasayat
    • bench:arijit pasayat
  • page_num (integer, optional, default: 0): Page index.
  • max_pages (integer, optional, default: 1): Maximum pages to retrieve.
  • max_cites (integer, optional): Optional list size for citations per matching document (capped at 50).

2. get_document

Retrieve the full text, metadata, and citation listings of a specific document.

• Arguments:
  • doc_id (integer, required): Unique document ID (tid).
  • max_cites (integer, optional, default: 0): Maximum citations to retrieve in citeList (up to 50).
  • max_citedby (integer, optional, default: 0): Maximum references citing this document in citedbyList (up to 50).

3. get_document_fragment

Retrieve highlighted text fragments/snippets matching a query within a specific document.

• Arguments:
  • doc_id (integer, required): Unique document ID (tid).
  • query (string, required): Search term or phrase to find fragments of.

4. get_original_document

Retrieve the original PDF/text copy if available (Base64-encoded PDF).

• Arguments:
  • doc_id (integer, required): Unique document ID (tid).