jupyter_server_ai_tools 0.1.2

No project description provided

🧠 jupyter-server-ai-tools

A Jupyter Server extension for discovering and aggregating callable tools from other extensions.

This project provides a structured way for extensions to declare tools using ToolDefinition objects, and for agents or other consumers to retrieve those tools — with optional metadata validation.

---

✨ Features

• ✅ Simple, declarative ToolDefinition API for registering callable tools
• ✅ Automatic metadata inference from Python function signature and docstring
• ✅ find_tools() for discovering tools from all installed Jupyter server extensions
• ✅ run_tools() for executing tools from structured call objects (supports sync, async, and multiple tool call formats)
• ✅ Built-in support for OpenAI, Anthropic, MCP, and Vercel tool call schemas
• ✅ Custom parser support for user-defined tool call formats
• ✅ Clean separation between tool metadata and callable execution
• ✅ Optional JSON Schema validation to enforce tool structure at definition time

---

📦 Install

pip install jupyter_server_ai_tools

To install for development:

git clone https://github.com/Abigayle-Mercer/jupyter-server-ai-tools.git
cd jupyter-server-ai-tools
pip install -e ".[lint,test]"

Usage

Expose tools in your own extensions:

from jupyter_server_ai_tools.models import ToolDefinition

def greet(name: str):
    """Say hello to someone."""
    return f"Hello, {name}!"

def jupyter_server_extension_tools():
    return [ToolDefinition(callable=greet)]

Discover tools from all extensions:

from jupyter_server_ai_tools.tool_registry import find_tools

tools = find_tools(extension_manager)

Execute tools via structured calls:

The run_tools() function allows dynamic execution of tool calls using a standard format such as "mcp", "openai", "anthropic", or "vercel":

from jupyter_server_ai_tools.tool_registry import run_tools

tool_calls = [
    {"name": "greet", "input": {"name": "Abigayle"}}
]

results = await run_tools(
    extension_manager=serverapp.extension_manager,
    tool_calls=tool_calls,
    parse_fn="mcp"
)

🧪 Running Tests

pip install -e ".[test]"
pytest

🧼 Linting and Formatting

pip install -e ".[lint]"
bash .github/workflows/lint.sh

Tool Output Example

Given the greet() tool above, find_tools(return_metadata_only=True) will return:

[
  {
    "name": "greet",
    "description": "Say hello to someone.",
    "inputSchema": {
      "type": "object",
      "properties": {
        "name": { "type": "string" }
      },
      "required": ["name"]
    }
  }
]

Impact

This system enables:

• Extension authors to register tools with minimal effort
• Agent builders to dynamically discover and bind tools
• Compatibility with multiple tool call formats, including OpenAI, Anthropic, MCP, and Vercel

🧹 Uninstall

pip uninstall jupyter_server_ai_tools