abcoder

Agentic backend coder - A Jupyter notebook manager with MCP (Model Context Protocol) integration for AI-assisted code execution and bioinformatics workflows.

Supports multiple Jupyter kernels (Python, R, etc.) for parallel notebook management.

🪩 What can it do?

• Jupyter Notebook Management: Create, switch between, and manage multiple Jupyter notebooks
• Multi-kernel Support: Manage and run code in multiple Jupyter kernels (e.g., Python, R) simultaneously
• Code Execution: Execute single-step or multi-step code in Jupyter kernels
• Variable Backup: Safely backup variables before code execution to prevent data loss
• API Documentation: Query function and API documentation directly from the kernel
• Output Handling: Capture and display execution results, errors, and generated figures
• Bioinformatics Integration: Designed for bioinformatics workflows with support for common libraries like scanpy, pandas, numpy, etc.

❓ Who is this for?

• Bioinformaticians who want AI assistance in their Jupyter workflows
• AI developers building agents that need to execute code in Jupyter environments
• Researchers who want to integrate AI tools with their computational notebooks
• Anyone who wants to use natural language to control Jupyter notebook execution

🌐 Where to use it?

You can use abcoder in most AI clients, plugins, or agent frameworks that support the MCP:

• AI clients: Cherry Studio, Claude Desktop, etc.
• Plugins: Cline, etc.
• Agent frameworks: Agno, etc.

🎬 Demo

A demo showing AI-assisted bioinformatics analysis in a Jupyter notebook using natural language commands through abcoder.

Click the image above to watch the demo video

📚 Documentation

For complete documentation, visit: https://github.com/huang-sh/abcoder

🏎️ Quickstart

Install

Install from PyPI:

pip install abcoder

Test the installation:

abcoder run

Configuration

Run abcoder locally

First, check the installation path:

which abcoder
# Example output: /home/user/bin/abcoder

Configure your MCP client:

{
  "mcpServers": {
    "abcoder": {
      "command": "/home/user/bin/abcoder",
      "args": ["run"]
    }
  }
}

Run abcoder remotely

Start the server on your remote machine:

abcoder run --transport shttp --port 8000

Configure your local MCP client:

{
  "mcpServers": {
    "abcoder": {
      "url": "http://localhost:8000/mcp"
    }
  }
}

🛠️ Available Tools

Notebook Management

• create_notebook: Create a new Jupyter notebook with specified ID and path
• switch_active_notebook: Switch between different notebooks

Code Execution

• single_step_execute: Execute a single code block
• multi_step_execute: Execute multiple code steps with cell addition
• query_api_doc: Query function documentation from the kernel

Features

• Variable Backup: Automatically backup variables before execution
• Error Handling: Comprehensive error capture and reporting
• Output Display: Support for text, images, and other display data
• Kernel Management: Automatic kernel lifecycle management

🤝 Contributing

If you have any questions, welcome to submit an issue, or contact me(hsh-me@outlook.com). Contributions to the code are also welcome!