notebook_nanoservice

Nanoservices for notebooks.

Quickly and effortlessly expose your notebook's functions via a REST API.

Note: This library is designed to scaffold a small backend service for ease of development and prototyping purposes only. Not for production use.

Features

• Hot functions - redefine your functions as you go without restarting the service.
• No decoration required.
• Auto serialization for primitive types, lists, Pandas dataframes, and images. Extensible for other types.
• Works with Jupyter & Spark notebooks, probably others.
• No dependency conflicts.
• API manifest in JSON or markdown, suitable for use with LLMs.

Scenarios

• Share your REST API via a tunnel such as Ngrok or Microsoft DevTunnels.
• Stand up an MCP server over your REST API.

Installation

pip install notebook-nanoservice

Usage

Initialize a server

from notebook_nanoservice import NanoService
service = NanoService(globals())
service.start()

View the manifest

View your API manifest in multiple formats:

• JSON: http://localhost:5001 (default lightweight format)
• Markdown: http://localhost:5001/?format=md
• OpenAPI v3: http://localhost:5001/?format=openapi

Invoke a function

If you have a function such as:

def concat(a, b):
    return a + b

It can be invoked at http://localhost:5001/api/concat?a=value1&b=value2

See test/sample.ipynb for examples of type casting the parameters.

Stop and free up the port

service.stop()

API

Constructor

The NanoService class is initialized to expose the global context of a Jupyter notebook or Python script via a REST API:

NanoService(global_context, host, port, include_source)

Parameters

• global_context (dict, required): The global context to expose. Typically passed as globals() from the notebook or script.
• host (str, optional): The host address for the server. Defaults to "127.0.0.1".
• port (int, optional): The port number for the server. Defaults to 5001.
• include_source (bool, optional): Whether to include the source code of functions in the metadata. Defaults to False.

Multiple service instances can run simultaneously by having unique port numbers.

# in a different notebook
from notebook_nanoservice import NanoService
service = NanoService(globals(), port=5002)
service.start()

ignore_functions

By default, all user-defined functions are exposed via the REST API. The ignore_functions property is a list of function names that the server will ignore when exposing the global context. By default, it includes functions like exit, quit, and others that are not meant to be exposed.

You can append to this list to exclude additional functions from being exposed. For example:

service.ignore_functions.append("my_function_to_ignore")

Contributing

Contributions are welcome! Please submit a pull request or open an issue on the GitHub repository.

Development

Install development dependencies:

pip install -r requirements.txt

Install the package locally: Open test/local_install.ipynb and follow instructions within.

License

This project is licensed under the MIT License. See the LICENSE file for details.