ai-tool-registry 0.7.4

Advanced tool registration system for Anthropic Claude integration with automatic schema generation and validation

Universal Tool Registry Module

Advanced tool registration system for multiple AI providers with automatic schema generation, validation, and error handling. Supports Anthropic Claude, OpenAI, Mistral AI, AWS Bedrock, and Google Gemini.

Features

• Multi-provider support - Works with all major AI providers
• Automatic JSON schema generation from function signatures
• Pydantic model integration and validation
• ToolContext parameter filtering - Automatic exclusion of context parameters with type safety
• Legacy parameter filtering for internal/context parameters
• Unified interface across different AI providers
• Comprehensive error handling and logging
• Type safety with full type hints
• Optional dependencies - Install only what you need

Installation

Basic Installation

# Using UV (recommended)
uv add ai-tool-registry

# Using pip
pip install ai-tool-registry

Provider-Specific Installation

# For Anthropic Claude
uv add ai-tool-registry[anthropic]

# For OpenAI
uv add ai-tool-registry[openai]

# For Mistral AI 
uv add ai-tool-registry[mistral]

# For AWS Bedrock
uv add ai-tool-registry[bedrock]

# For Google Gemini
uv add ai-tool-registry[gemini]

# Install all providers
uv add ai-tool-registry[all]

Quick Start

from tool_registry_module import tool, build_registry_openai, build_registry_anthropic, ToolContext
from pydantic import BaseModel
from typing import Annotated


class UserData(BaseModel):
    name: str
    age: int


@tool(description="Process user information")
def process_user(
    input: UserData, 
    context: ToolContext[dict] = None  # Automatically excluded from schema
) -> UserData:
    # context parameter is available for use but won't appear in AI tool schema
    return input


# Build registries for different providers
openai_registry = build_registry_openai([process_user])
anthropic_registry = build_registry_anthropic([process_user])

# Use with respective APIs
openai_tools = [entry["representation"] for entry in openai_registry.values()]
anthropic_tools = [entry["representation"] for entry in anthropic_registry.values()]

Multi-Provider Examples

OpenAI Function Calling

from tool_registry_module import tool, build_registry_openai
import openai

@tool(description="Get weather information")
def get_weather(location: str, unit: str = "celsius") -> str:
    return f"Weather in {location}: 22°{unit[0].upper()}"

# Build OpenAI registry
registry = build_registry_openai([get_weather])
tools = [entry["representation"] for entry in registry.values()]

# Use with OpenAI
client = openai.OpenAI()
response = client.chat.completions.create(
    model="gpt-4",
    messages=[{"role": "user", "content": "What's the weather in Paris?"}],
    tools=tools
)

Anthropic Claude

from tool_registry_module import tool, build_registry_anthropic
import anthropic

registry = build_registry_anthropic([get_weather])
tools = [entry["representation"] for entry in registry.values()]

# Use with Anthropic
client = anthropic.Anthropic()
response = client.messages.create(
    model="claude-3-sonnet-20240229",
    max_tokens=1000,
    messages=[{"role": "user", "content": "What's the weather in Paris?"}],
    tools=tools
)

AWS Bedrock

from tool_registry_module import tool, build_registry_bedrock
import boto3

registry = build_registry_bedrock([get_weather])
tools = [entry["representation"] for entry in registry.values()]

# Use with Bedrock
client = boto3.client("bedrock-runtime")
# Use tools in your Bedrock converse API calls

Google Gemini

from tool_registry_module import tool, build_registry_gemini
import google.generativeai as genai

registry = build_registry_gemini([get_weather])
tools = [entry["representation"] for entry in registry.values()]

# Use with Gemini
model = genai.GenerativeModel('gemini-pro')
# Use tools in your Gemini function calling

Mistral AI

from tool_registry_module import tool, build_registry_mistral
from mistralai.client import MistralClient

registry = build_registry_mistral([get_weather])
tools = [entry["representation"] for entry in registry.values()]

# Use with Mistral
client = MistralClient()
# Use tools in your Mistral function calling

Advanced Usage

Parameter Filtering

Using ToolContext (Recommended)

Use ToolContext to mark parameters that should be automatically excluded from schemas:

from tool_registry_module import tool, ToolContext
from typing import Annotated

@tool(description="Process user data with context")
def process_data(
    user_input: str,
    context: ToolContext[dict],  # Direct ToolContext generic
    session: Annotated[str, ToolContext],  # Annotated ToolContext
    debug_flag: bool = False
) -> str:
    # context and session parameters are automatically excluded from the schema
    # but available for use in your function
    return f"Processed: {user_input}"

ToolContext Features:

• Automatic exclusion - No need to manually specify ignore_in_schema
• Type safety - Full type hints with generic support
• Reference preservation - Objects maintain their references for mutation within functions
• Multiple forms - Both direct (ToolContext[T]) and annotated (Annotated[T, ToolContext]) syntax
• Union validation - Prevents incorrect usage in union types

Supported ToolContext patterns:

# Direct generic types
param1: ToolContext[dict]
param2: ToolContext[str] 

# Annotated types
param3: Annotated[str, ToolContext]
param4: Annotated[dict, ToolContext, "description"]

# Union types will raise TypeError (prevented for safety)
# param5: Union[str, ToolContext[dict]]  # ❌ Not allowed

Reference Preservation Example:

from tool_registry_module import tool, ToolContext

# Context objects maintain their references for mutation
@tool(description="Track user interactions")
def track_interaction(
    action: str,
    user_context: ToolContext[dict]  # This dict can be modified
) -> str:
    # Modify the context object - changes persist outside function
    user_context["actions"] = user_context.get("actions", [])
    user_context["actions"].append(action)
    user_context["last_action"] = action
    return f"Tracked: {action}"

# Usage
context = {"user_id": "123"}
track_interaction("login", user_context=context)
track_interaction("view_profile", user_context=context)

# Context object is modified:
print(context)  
# {'user_id': '123', 'actions': ['login', 'view_profile'], 'last_action': 'view_profile'}

Legacy Parameter Filtering

You can still manually exclude parameters using ignore_in_schema:

@tool(
    description="Calculate area with debug output",
    ignore_in_schema=["debug_mode", "context"]
)
def calculate_area(length: float, width: float, debug_mode: bool = False, context: str = "calc") -> float:
    if debug_mode:
        print(f"Calculating area for {length} x {width}")
    return length * width

Cache Control (Anthropic)

Add cache control for better performance with Anthropic:

@tool(
    description="Expensive computation",
    cache_control={"type": "ephemeral"}
)
def expensive_function(data: str) -> str:
    # Expensive computation here
    return processed_data

Registry Utilities

from tool_registry_module import get_tool_info, validate_registry

# Get information about a specific tool
info = get_tool_info(registry, "process_user")
print(info["description"])

# Validate registry structure
is_valid = validate_registry(registry)

Tool Use Handling

The registry is a dictionary that enables dynamic function calling for AI tool responses:

from tool_registry_module import tool, build_registry_anthropic

@tool(description="Add two numbers")
def add_numbers(a: int, b: int) -> int:
    return a + b

@tool(description="Get weather info")
def get_weather(city: str, units: str = "celsius") -> str:
    return f"Weather in {city}: 22°{units[0].upper()}"

# Build registry
registry = build_registry_anthropic([add_numbers, get_weather])

# Handle tool use responses dynamically
def handle_tool_calls(tool_calls, registry):
    results = []
    for tool_call in tool_calls:
        tool_name = tool_call.name
        tool_args = tool_call.input
        
        if tool_name in registry:
            try:
                # Get function from registry and execute
                tool_func = registry[tool_name]["tool"]
                result = tool_func(**tool_args)
                results.append({
                    "tool_use_id": tool_call.id,
                    "content": str(result)
                })
            except Exception as e:
                results.append({
                    "tool_use_id": tool_call.id,
                    "error": f"Error: {e}"
                })
        else:
            results.append({
                "tool_use_id": tool_call.id,
                "error": f"Tool '{tool_name}' not found"
            })
    
    return results

# Registry structure: {tool_name: {"tool": callable, "representation": provider_format}}
# Use registry[tool_name]["tool"] for dynamic function calling

Supported Providers

Provider	Function	Format
Anthropic Claude	build_registry_anthropic()	Claude ToolParam
OpenAI	build_registry_openai()	OpenAI Function Call
Mistral AI	build_registry_mistral()	Mistral Function Call
AWS Bedrock	build_registry_bedrock()	Bedrock ToolSpec
Google Gemini	build_registry_gemini()	Gemini FunctionDeclaration

Requirements

• Python 3.12+
• pydantic >= 2.0.0 (required)

Optional Provider Dependencies

• anthropic >= 0.52.2 (for Anthropic Claude)
• openai >= 1.0.0 (for OpenAI)
• mistralai >= 0.4.0 (for Mistral AI)
• boto3 >= 1.34.0 (for AWS Bedrock)
• google-generativeai >= 0.3.0 (for Google Gemini)

Migration from v2.x

The old build_registry_anthropic_tool_registry() function is still available for backward compatibility but deprecated. Use build_registry_anthropic() instead.

License

MIT License

Development

Setup

# Clone and install with dev dependencies
git clone https://github.com/kazmer97/ai-tool-registry.git
cd ai-tool-registry
uv sync --extra dev

# Run linting
uv run ruff check .
uv run ruff format .

Testing

# Run tests (when available)
uv run pytest

# Type checking
uv run mypy tool_registry_module/

Contributing

1. Fork the repository
2. Create a feature branch (git checkout -b feature/amazing-feature)
3. Run linting: uv run ruff check . && uv run ruff format .
4. Commit changes (git commit -m 'Add amazing feature')
5. Push to branch (git push origin feature/amazing-feature)
6. Open a Pull Request