Mcp Python Code Interpreter

MCP Python Code Interpreter

A stateful Python code execution environment with Jupyter kernel support, built on the Model Context Protocol (MCP). Execute Python code with persistent state, automatic visualization handling, and secure session management.

Features

• Stateful Execution: Variables, imports, and state persist across multiple code executions within a session
• Jupyter Kernel Backend: Full Jupyter kernel support with IPython features
• Session Management: Secure session IDs with automatic 30-minute timeout
• Visualization Support: Automatic capture and export of matplotlib, seaborn, and Plotly figures
• File Generation: Access generated files (images, data files, plots) via MCP resources
• Scientific Computing: Pre-configured with pandas, numpy, matplotlib, seaborn, plotly, and sympy
• Automatic Cleanup: Background task removes expired sessions and orphaned files

Installation

Using Docker (Recommended)

# Build the image
docker build -t mcp-python-interpreter .

# Run the container
docker run -p 8000:8000 mcp-python-interpreter

Local Installation

# Create virtual environment
python -m venv .venv
source .venv/bin/activate  # On Windows: .venv\Scripts\activate

# Install dependencies
pip install -r requirements.txt

# Run the server
python server.py

Usage

The server exposes three tools and a resource endpoint for file access.

1. Execute Code

Execute Python code in a persistent Jupyter kernel environment.

First execution (creates new session):

{
  "code": "import pandas as pd\nx = 42\nprint('Hello, World!')",
  "session_id": ""  # Empty string or "0" for new session
}

Response:

{
  "success": true,
  "output": ["Hello, World!\n"],
  "result": null,
  "session_info": {
    "session_id": "abc123xyz456",
    "instructions": "Use this `session_id` in subsequent requests..."
  }
}

Subsequent executions (reuse session):

{
  "code": "print(x * 2)  # Variable persists from previous execution",
  "session_id": "abc123xyz456"
}

2. Create Visualizations

Matplotlib, seaborn, and Plotly figures are automatically captured and saved.

{
  "code": """
import matplotlib.pyplot as plt
import numpy as np

x = np.linspace(0, 10, 100)
plt.plot(x, np.sin(x))
plt.title('Sine Wave')
plt.savefig('sine_wave.png')
plt.show()
""",
  "session_id": "abc123xyz456"
}

Response includes file references:

{
  "success": true,
  "output": ["📊 File created: sine_wave.png (image/png, 45231 bytes)"],
  "files": [
    {
      "uri": "kernel://abc123xyz456/sine_wave.png",
      "mime_type": "image/png",
      "name": "sine_wave.png",
      "size": 45231
    }
  ]
}

3. List Session Files

{
  "session_id": "abc123xyz456"
}

4. Clear Session

Manually remove a session and all its files:

{
  "session_id": "abc123xyz456"
}

5. Access Files

Files are accessed via the MCP resource protocol:

kernel://{session_id}/{filename}

Example: kernel://abc123xyz456/sine_wave.png

Session Management

Session Lifecycle

1. Creation: First call with empty session_id generates a secure 16-character ID
2. Active: Session remains active while being used
3. Timeout: Sessions expire after 30 minutes of inactivity
4. Cleanup: Expired sessions are automatically removed (background task runs every 5 minutes)

Session Expiration

If you try to use an expired session, you'll receive a SessionExpiredError:

{
  "success": false,
  "error": "SessionExpiredError: Session abc123xyz456 not found or has expired...",
  "traceback": []
}

Solution: Create a new session and re-execute your setup code.

Available Libraries

The following libraries are pre-installed:

• Data Science: pandas, numpy
• Visualization: matplotlib, seaborn, plotly, kaleido
• Mathematics: sympy
• Jupyter: ipykernel, jupyter-client

Additional packages can be installed at runtime using !pip install within your code.

Configuration

Environment Variables

• LOG_LEVEL: Logging level (default: INFO)
• JUPYTER_PLATFORM_DIRS: Set to 1 (suppresses Jupyter warnings)

Server Configuration

Edit config.py to modify:

NOTEBOOKS_FOLDER = ROOT_DIR / 'notebooks'  # Session storage location
KERNEL_TIMEOUT = 10  # Kernel operation timeout (seconds)

Edit server.py to modify:

SESSION_TIMEOUT = 30 * 60  # Session expiration (seconds)

API Reference

Tools

execute_code

Execute Python code in a persistent Jupyter kernel environment.

Parameters:

• code (str): Python code to execute (multi-line supported)
• session_id (str, optional): Session identifier (empty or "0" for new session)

Returns:

• success (bool): Execution status
• output (list): stdout/stderr text
• result (str|None): Last expression value
• error (str|None): Error message if failed
• traceback (list): Full traceback if error
• files (list): File references with URIs
• session_info (dict|None): Session info for new sessions

list_session_files

List all files generated in a session.

Parameters:

• session_id (str): Session identifier

Returns:

• session_id (str): The session ID
• files (list): List of file references
• error (str): Error message if session not found

clear_session

Manually clear a session and shutdown its kernel.

Parameters:

• session_id (str): Session identifier to clear

Returns:

• success (bool): Operation status
• message (str): Status message

Resources

kernel://{session_id}/{filename}

Retrieve file content from a session.

URI Format: kernel://session_id/filename

Returns: File content (binary for images, text for text files)

Architecture

┌─────────────────┐
│  MCP Client     │
│  (Claude, etc)  │
└────────┬────────┘
         │
         │ MCP Protocol
         │
┌────────▼────────┐
│    MCP Server   │
│  (server.py)    │
└────────┬────────┘
         │
         │ Manages
         │
┌────────▼────────┐
│ Session Manager │
│ - Create/Track  │
│ - Cleanup       │
└────────┬────────┘
         │
         │ Controls
         │
┌────────▼────────┐
│ Jupyter Kernels │
│ (per session)   │
└─────────────────┘

Development

Project Structure

mcp-python-code-interpreter/
├── config.py          # Configuration and initialization
├── models.py          # Pydantic data models
├── notebook.py        # Jupyter kernel management
├── server.py          # MCP server and tools
├── Dockerfile         # Docker configuration
├── requirements.txt   # Python dependencies
└── notebooks/         # Session storage (auto-created)

Source & license

This open-source MCP server is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: khshanovskyi
• Source: khshanovskyi/mcp-python-code-interpreter
• License: MIT

Install and usage instructions live in the source repository linked above.