Bangladesh Law MCP Server

Bangladesh Law MCP Server

A Model Context Protocol server that lets anyAI assistant — Claude, Gemini, GPT, Cursor, VS Code, Continue.dev,or any MCP-compatible client — query the full text of Bangladeshi legislationdirectly from natural language.

Built on top of theBangladesh Legal Acts Dataset,which contains 1,484+ acts scraped from http://bdlaws.minlaw.gov.bd/ andenriched with the historical government and legal-system context for eachact's period.

💚 Credits

This project was made possible by the generous help of two friends:

• Rubaiyat — for the early feedback, the test queries that shaped thetool surface, and the patient bug reports that turned half-brokenprototypes into something that actually works end-to-end.
• Jarif — for helping debug the stdio transport, walking throughMCP-client integration on Windows, and pushing for the polished READMEand CI setup that make this thing actually shippable.

Thank you both — this project exists because of you. 💚

And of course, thanks to sakhadib and everycontributor to the originalBangladesh Legal Acts Dataset —without that carefully curated corpus, none of this would exist.

✨ What you can ask

Once connected, your AI assistant can answer questions like:

• "Find all Bangladesh acts from the 1970s related to land."
• "What was the legal framework when the Penal Code 1860 was enacted?"
• "Show me section 3 of the Code of Criminal Procedure."
• "Which government was in power when the Bangladesh University of Professionals Act 2009 was passed?"
• "Search the entire corpus for acts that mention 'environment protection'."
• "Is the Companies Act 1994 still in force?"

🧰 Tools exposed

Each act is also exposed as an MCP resource at act://<act-id> so clientsthat prefer resource-style reads can pull the JSON directly.

📦 Installation

1. Clone this repo and the dataset side-by-side

git clone https://github.com/sakhadib/bangladesh-law-mcp.git
git clone https://github.com/sakhadib/Bangladesh-Legal-Acts-Dataset.git

You should now have:

bangladesh-law-mcp/
Bangladesh-Legal-Acts-Dataset/

The server will auto-discover the dataset in this sibling layout. If you keepthe dataset anywhere else, set BLA_DATA_DIR to its Data/acts folder (or tothe parent Data/ folder).

About Git LFS: the dataset uses LFS for the full corpus. If you only seea few hundred .json files in Bangladesh-Legal-Acts-Dataset/Data/acts/,run git lfs install && git lfs pull inside the dataset repo.

2. Install Python dependencies

cd bangladesh-law-mcp
python -m venv .venv
# Windows
.venv\Scripts\activate
# macOS / Linux
source .venv/bin/activate

pip install -r requirements.txt

Or install the package itself:

pip install .

This registers a bangladesh-law-mcp command on your PATH.

3. Sanity check

python server.py

The server should print something like:

[INFO] bangladesh-law-mcp: Loading acts from .../Bangladesh-Legal-Acts-Dataset/Data/acts
[INFO] bangladesh-law-mcp: Indexing 1484 act files
[INFO] bangladesh-law-mcp: Index built: 1484 acts

and then sit silently on stdout (it's waiting for an MCP client on stdin). Press Ctrl+C to stop.

🔌 Connect an AI client

Pick your client. All of them connect over stdio — no API key, no network.

Claude Desktop

Edit claude_desktop_config.json:

• Windows: %APPDATA%\Claude\claude_desktop_config.json
• macOS: ~/Library/Application Support/Claude/claude_desktop_config.json
• Linux: ~/.config/Claude/claude_desktop_config.json

{
  "mcpServers": {
    "bangladesh-law": {
      "command": "python",
      "args": ["C:\\absolute\\path\\to\\bangladesh-law-mcp\\server.py"]
    }
  }
}

Restart Claude Desktop. You'll see a 🔨 icon with the new tools in the chatcomposer.

VS Code / Cursor (one-click via .vscode/mcp.json)

A starter config is included at .vscode/mcp.json (edit the absolute pathto match your machine):

{
  "servers": {
    "bangladesh-law": {
      "type": "stdio",
      "command": "python",
      "args": ["C:\\absolute\\path\\to\\bangladesh-law-mcp\\server.py"]
    }
  }
}

Then in VS Code run MCP: List Servers from the command palette and startbangladesh-law.

Gemini (via an MCP-compatible adapter)

Gemini itself doesn't speak MCP natively, but open-source MCP bridges(LangChain's MCP adapter, the MCP Inspector, etc.) let you expose MCP tools toGemini. After installing such an adapter, point it at:

python /absolute/path/to/bangladesh-law-mcp/server.py

GPT / ChatGPT (via MCP plugin / LangChain)

Same story — wire the stdio server into any MCP-aware orchestrator(LangChain's MCPAdapter, AutoGen, CrewAI, etc.). The tool surface describedabove is what your GPT agent will see.

Continue.dev / Cline / Roo Code

Add to your MCP config:

{
  "mcpServers": [
    {
      "name": "bangladesh-law",
      "command": "python",
      "args": ["/absolute/path/to/bangladesh-law-mcp/server.py"]
    }
  ]
}

🌐 Live Hosted Instance (Fly.io)

The server is deployed as an always-on Streamable HTTP endpoint onFly.io. Any MCP client that supports remote serverscan connect to:

https://bangladesh-law-mcp.fly.dev/mcp

A liveness check is available at:

https://bangladesh-law-mcp.fly.dev/healthz

Connecting a remote-capable client

VS Code / Cursor (Streamable HTTP):

{
  "servers": {
    "bangladesh-law": {
      "type": "streamableHttp",
      "url": "https://bangladesh-law-mcp.fly.dev/mcp"
    }
  }
}

Claude Desktop (if/when remote MCP is supported):

{
  "mcpServers": {
    "bangladesh-law": {
      "type": "streamableHttp",
      "url": "https://bangladesh-law-mcp.fly.dev/mcp"
    }
  }
}

Note: Most desktop MCP clients today only support stdio. Thehosted endpoint is primarily for web-based agents, custom integrations,and future clients that support remote MCP. For local use, the stdiotransport (above) remains the simplest option.

Deploying your own instance

1. Install flyctl and sign up.

2. Push this repo to GitHub.

3. Create a Fly app:

   fly launch --no-deploy   # generates fly.toml, uses the Dockerfile
   

4. Deploy:

   fly deploy --remote-only
   

5. (Optional) Set up auto-deploy on push — add a FLY_API_TOKEN GitHubsecret (created with fly tokens create deploy) and the included.github/workflows/deploy.yml will deploy on every push to main.

⚙️ Configuration

🛠 Development

Run a one-liner smoke test (no client needed):

python -c "import asyncio, server
async def m():
    out = await server.mcp.call_tool('get_statistics', {})
    print(out[1])
asyncio.run(m())"

Lint:

pyflakes server.py

The included GitHub Actions workflow (.github/workflows/ci.yml) lints andbyte-compiles on Python 3.10, 3.11, and 3.12 for every push and PR.

🗂 Project layout

bangladesh-law-mcp/
├── .dockerignore
├── .github/
│   ├── ISSUE_TEMPLATE/
│   │   ├── bug_report.md
│   │   └── feature_request.md
│   └── workflows/
│       ├── ci.yml
│       └── deploy.yml        # auto-deploy to Fly.io
├── .gitignore
├── Dockerfile                # multi-stage build (clones dataset with LFS)
├── LICENSE
├── README.md
├── SECURITY.md
├── fly.toml                  # Fly.io app config (always-on free tier)
├── pyproject.toml
├── requirements.txt
└── server.py                 # the entire MCP server

🙏 Thanks

The full credit list lives at the top of this README in the💚 Credits section — please scroll up.

📄 License

• The server code in this repository is released under theMIT License.
• The underlying data is fromsakhadib/Bangladesh-Legal-Acts-Datasetand is licensed underCC BY 4.0. Please attributethe original authors when reusing the data.