license: mit title: Tensorus MCP sdk: python emoji: 🐠 colorFrom: blue colorTo: yellow short_description: Model Context Protocol server and client for Tensorus tensor database

Tensorus MCP

Model Context Protocol (MCP) server and client for Tensorus tensor database operations. This package provides a standardized interface for AI agents and LLMs to interact with Tensorus capabilities using the Model Context Protocol.

Related MCP server: Directus MCP Server

Features

• MCP Server: Python implementation using fastmcp for tensor database operations

• MCP Client: Python client library for easy integration with MCP servers

• Tensor Operations: Complete set of tensor manipulation tools via MCP

• Dataset Management: Create, list, and manage tensor datasets

• Demo Mode: Pre-configured mock data for testing and demonstration

Installation

pip install fastmcp
pip install -r requirements.txt

Quick Start

Starting the MCP Server

python -m tensorus_mcp.server

For web endpoint support:

python -m tensorus_mcp.server --transport streamable-http

Demo Mode

For demonstration or testing purposes, run the server in demo mode:

python -m tensorus_mcp.server --demo-mode

Using the Python Client

from tensorus_mcp.client import TensorusMCPClient

async def example():
    async with TensorusMCPClient.from_http("http://localhost:8000/mcp/") as client:
        # List available datasets
        datasets = await client.list_datasets()
        print(f"Available datasets: {datasets}")
        
        # Create a new dataset
        await client.create_dataset("my_dataset")
        
        # Ingest a tensor
        result = await client.ingest_tensor(
            dataset_name="my_dataset",
            tensor_shape=[2, 2],
            tensor_dtype="float32",
            tensor_data=[[1.0, 2.0], [3.0, 4.0]],
            metadata={"source": "example"}
        )
        print(f"Ingested tensor with ID: {result['record_id']}")

MCP Demo Script

Prerequisites

• Tensorus MCP Server running (python -m tensorus_mcp.server)

• For live mode: Tensorus backend API accessible

• For demo mode: No additional setup required

Demo Scenario: MCP Client Interaction

Goal: Demonstrate how an external AI agent can leverage Tensorus via MCP.

1. Start MCP Server:

   python -m tensorus_mcp.server --demo-mode

2. Connect via Python Client:

   from tensorus_mcp.client import TensorusMCPClient
   
   async def demo():
       async with TensorusMCPClient.from_http("http://localhost:8000/mcp/") as client:
           # List available datasets
           datasets = await client.list_datasets()
           print(f"Available datasets: {datasets}")
           
           # Create a new dataset
           result = await client.create_dataset("demo_dataset")
           print(f"Created dataset: {result}")
           
           # Ingest sample tensor data
           tensor_result = await client.ingest_tensor(
               dataset_name="demo_dataset",
               tensor_shape=[3, 3],
               tensor_dtype="float32",
               tensor_data=[[1.0, 2.0, 3.0], [4.0, 5.0, 6.0], [7.0, 8.0, 9.0]],
               metadata={"source": "mcp_demo", "type": "sample_matrix"}
           )
           print(f"Ingested tensor: {tensor_result}")
           
           # Apply tensor operation (transpose)
           op_result = await client.apply_operation(
               operation="transpose",
               dataset_name="demo_dataset",
               record_id=tensor_result["record_id"],
               dim0=0,
               dim1=1
           )
           print(f"Applied transpose operation: {op_result}")

3. Conceptual Client Interaction (JavaScript):

   // Example of how other AI agents could interact via MCP
   async function mcpDemo() {
       // List available tools
       const { tools } = await client.request({ method: 'tools/list' }, {});
       console.log("Available Tensorus Tools:", tools.map(t => t.name));
       
       // Create dataset via MCP
       const createResponse = await client.request({ method: 'tools/call' }, {
           name: 'tensorus_create_dataset',
           arguments: { dataset_name: 'mcp_demo_dataset' }
       });
       console.log("Dataset created:", createResponse.content[0].text);
       
       // Ingest tensor via MCP
       const ingestResponse = await client.request({ method: 'tools/call' }, {
           name: 'tensorus_ingest_tensor',
           arguments: {
               dataset_name: 'mcp_demo_dataset',
               tensor_shape: [2, 2],
               tensor_dtype: 'float32',
               tensor_data: [[1.0, 2.0], [3.0, 4.0]],
               metadata: { source: 'mcp_demo' }
           }
       });
       console.log("Tensor ingested:", ingestResponse.content[0].text);
   }

Available MCP Tools

Dataset Management

• tensorus_list_datasets: Lists all available datasets

• tensorus_create_dataset: Creates a new dataset

• tensorus_delete_dataset: Deletes an existing dataset

Tensor Operations

• tensorus_ingest_tensor: Ingests a new tensor into a dataset

• tensorus_get_tensor_details: Retrieves tensor data and metadata

• tensorus_delete_tensor: Deletes a specific tensor

• tensorus_update_tensor_metadata: Updates tensor metadata

Tensor Computations

• tensorus_apply_unary_operation: Operations like log, reshape, transpose, sum, mean

• tensorus_apply_binary_operation: Operations like add, subtract, multiply, matmul

• tensorus_apply_list_operation: Operations like concatenate and stack

• tensorus_apply_einsum: Einstein summation operations

Diagnostic Tools

• mcp_server_status: Check server operational status

• connection_test: Lightweight connectivity check

• backend_ping: Test backend API health endpoint

• backend_connectivity_test: Verify backend communication

Configuration

API Key Management

When not in demo mode, provide authentication via:

1. Global API Key: Set when starting the server

   python -m tensorus_mcp.server --mcp-api-key YOUR_API_KEY

2. Per-Tool API Key: Pass api_key parameter in tool calls

Environment Variables

• TENSORUS_API_BASE_URL: Backend API URL (default: https://tensorus-core.hf.space)

• TENSORUS_MINIMAL_IMPORT: Set to 1 for lightweight imports

Demo Examples

Interactive Notebook

See examples/demo_notebook.ipynb for a complete interactive example.

Streamlit App

Launch the demo Streamlit app:

streamlit run examples/demo_app.py

Development

Running Tests

# Install test dependencies
pip install -r examples/requirements.txt

# Run MCP-specific tests
pytest tests/test_mcp_integration.py

Project Structure

tensorus_mcp/
├── __init__.py          # Package initialization
├── server.py            # MCP server implementation
├── client.py            # MCP client library
└── config.py            # Configuration management

examples/
├── demo_app.py          # Streamlit demo application
├── demo_notebook.ipynb  # Interactive Jupyter notebook
└── requirements.txt     # Demo dependencies

tests/
└── test_mcp_integration.py  # Integration tests

Usage in Claude Desktop

Add to your Claude Desktop MCP settings:

{
  "mcpServers": {
    "tensorus": {
      "command": "python",
      "args": ["-m", "tensorus_mcp.server"],
      "env": {
        "TENSORUS_API_BASE_URL": "https://tensorus-core.hf.space"
      }
    }
  }
}

API Reference

TensorusMCPClient Methods

• list_datasets(): Get all available datasets

• create_dataset(name, schema=None): Create a new dataset

• ingest_tensor(dataset_name, tensor_shape, tensor_dtype, tensor_data, metadata): Add tensor to dataset

• get_tensor_details(dataset_name, record_id): Retrieve tensor information

• apply_operation(operation, dataset_name, record_id, **kwargs): Apply tensor operations

Contributing

Contributions are welcome! Please feel free to open issues or submit pull requests.

License

MIT License