B2A Toolkit 🚀

A lightweight framework to expose any API as an agent-compatible tool. B2A Toolkit makes it easy to create tools that can be used by AI agents, with built-in support for OpenAI's MCP spec, LangChain, and CrewAI.

⭐️ If you find this toolkit useful, please consider giving it a star on GitHub! ⭐️

Installation

pip install b2a-toolkit

Quick Start

1. Define a tool using the @define_tool decorator:

from toolkit import define_tool, ToolAuth

@define_tool(
    name="create_expense",
    description="Creates an expense in the system",
    inputs={
        "amount": "number",
        "category": "string",
        "description": "string",
        "date": "string"
    },
    output_type="object",
    output_description="The created expense object",
    auth=ToolAuth(
        type="oauth",
        required=True,
        scopes=["expenses:write"]
    )
)
def create_expense(amount: float, category: str, description: str, date: str):
    # Your implementation here
    return {"id": "exp_123", "amount": amount, ...}

2. Build the tool manifest:

toolkit build examples/expense_tool.py -o manifest.json

3. Test your tool:

toolkit test create_expense examples/expense_tool.py

4. Start the tool server:

toolkit serve examples/expense_tool.py

This starts a local server with:

• GET /manifest.json - Get tool definitions
• POST /run/<tool_name> - Execute a tool

5. Set up webhooks for monitoring:

# Add a webhook for all tools
toolkit webhook add https://your-server.com/webhook

# Or add tool-specific webhook with signature verification
toolkit webhook add https://your-server.com/webhook --tool create_expense --secret your-secret

# List registered webhooks
toolkit webhook list

6. Monitor your tools:

# View recent tool calls
toolkit logs --limit 10

# Follow logs in real-time
toolkit logs --follow

# Filter logs for specific tool
toolkit logs --tool create_expense

# Launch web dashboard
toolkit dashboard

7. Replay and debug tool calls:

# Get a call ID from logs
toolkit logs --tool create_expense --limit 1

# Replay that specific call
toolkit replay <CALL_ID> examples/expense_tool.py

Tool Definition

The @define_tool decorator takes the following parameters:

• name: Name of the tool
• description: Description of what the tool does
• inputs: Dictionary mapping input parameter names to their types
• output_type: Type of the tool's output
• output_description: Description of the tool's output
• auth: Optional authentication configuration
• version: Version of the tool (default: "0.1.0")
• tags: Optional list of tags/categories

CLI Commands

Build Manifest

toolkit build <module_path> [--output <output_file>] [--format <json|openapi>]

Start Server

toolkit serve <module_path> [--host HOST] [--port PORT]

Test Tool

toolkit test <tool_name> <module_path>

View Logs

# View recent logs
toolkit logs [--tool TOOL_NAME] [--limit N] [--format text|json]

# Follow logs in real-time
toolkit logs --follow

# Inspect specific tool usage
toolkit inspect TOOL_NAME --last 5

Manage Webhooks

# Add a webhook
toolkit webhook add <URL> [--tool TOOL_NAME] [--secret SECRET]

# List registered webhooks
toolkit webhook list

# Remove a webhook
toolkit webhook remove <URL> [--tool TOOL_NAME]

Webhooks receive events:

• tool.call: When a tool execution starts
• tool.success: When a tool execution succeeds
• tool.error: When a tool execution fails

Example webhook payload:

{
  "event": "tool.success",
  "tool": "calculator",
  "timestamp": "2024-03-14T12:34:56.789Z",
  "data": {
    "inputs": {
      "x": 10,
      "y": 5,
      "operation": "add"
    },
    "outputs": {
      "result": 15
    }
  }
}

Replay Tool Calls

# Replay a specific tool call using its ID
toolkit replay <CALL_ID> <module_path>

The replay feature allows you to:

• Re-execute previous tool calls
• Compare original and replay outputs
• Debug tool behavior changes
• Verify tool consistency

Launch Dashboard

toolkit dashboard [--host HOST] [--port PORT]

Examples

Check out the examples/ directory for more examples:

• expense_tool.py: A simple expense creation tool
• More examples coming soon...

Features

• Tool Definition

  • Simple decorator-based API
  • Input/output validation
  • Type checking and conversion
  • Built-in documentation support

• Tool Server

  • Local HTTP server for tool execution
  • Automatic manifest generation
  • RESTful endpoints for tool invocation
  • Built-in validation and error handling

• Authentication

  • OAuth support
  • API key support
  • Configurable scopes
  • Optional auth requirements

• Framework Support

  • OpenAI MCP spec compatibility
  • LangChain tool format
  • CrewAI integration

• Tool Observability 🔍

  • Per-call logging with unique IDs
  • Real-time log streaming
  • Web dashboard for monitoring
  • Usage analytics and debugging

Contributing

Contributions are welcome! Please feel free to submit a Pull Request.

License

This project is licensed under the Apache License 2.0 - see the LICENSE file for details.