aw_agents

aw_agents - AI Agents for Agentic Workflows

A collection of AI agents that can be easily deployed to multiple chatbot platforms (Claude, ChatGPT, etc.) through adapters.

Agents:

• DownloadAgent: Smart content downloader with context-aware naming

Adapters:

• MCPAdapter: For Claude Desktop (MCP protocol)

• OpenAPIAdapter: For ChatGPT Custom GPTs (OpenAPI/FastAPI)

Example

>>> from aw_agents.agents.download import DownloadAgent
>>> from aw_agents.adapters import MCPAdapter
>>>
>>> agent = DownloadAgent()
>>> adapter = MCPAdapter(agent, "download-agent")
>>> # adapter.run() serves the agent via MCP for Claude (blocks; not run here)

class aw_agents.AgentBase

Base class for all AW agents.

Agents should inherit from this and implement the core methods. Adapters (MCP, OpenAPI) will wrap these agents for deployment.

abstractmethod execute_tool(name: str, arguments: Dict[str, Any]) → Dict[str, Any]

Execute a tool with given arguments.

Parameters:

• name – Tool name

• arguments – Tool arguments

Returns:

• success: bool

• data: Any

• message: Optional[str]

• warnings: Optional[list[str]]

Return type:

Result dictionary with at least

get_metadata() → Dict[str, Any]

Get agent metadata.

Returns:

Metadata including name, version, description

abstractmethod get_tools() → list[Dict[str, Any]]

Return tool definitions for this agent.

Each tool should have: - name: str - description: str - parameters: dict (JSON schema)

Returns:

List of tool definitions

class aw_agents.DownloadAgent(default_download_dir: str | Path | None = None, **kwargs)

Smart download agent with context-aware naming and intelligent link handling.

Features: - Detects landing pages and finds actual download links - Special handling for GitHub, HuggingFace, Kaggle - Context-aware file naming - Multiple content type support

>>> agent = DownloadAgent()
>>> tools = agent.get_tools()
>>> len(tools) >= 3
True

execute_tool(name: str, arguments: Dict[str, Any]) → Dict[str, Any]

Execute a tool with given arguments.

get_metadata() → Dict[str, Any]

Get agent metadata.

get_tools() → List[Dict[str, Any]]

Return tool definitions for this agent.

class aw_agents.MCPAdapter(agent: AgentBase, server_name: str)

Adapter to expose an AgentBase as an MCP server.

Usage:

agent = YourAgent() adapter = MCPAdapter(agent, server_name=”your-agent”) adapter.run()

async run()

Run the MCP server.

run_sync()

Run the MCP server (synchronous wrapper).

class aw_agents.OpenAPIAdapter(agent: AgentBase, *, title: str = 'AI Agent API', description: str = 'AI Agent exposed via OpenAPI', version: str = '1.0.0', server_url: str | None = None)

Adapter to expose an AgentBase as a FastAPI/OpenAPI service.

Usage:

agent = YourAgent() adapter = OpenAPIAdapter(agent, title=”Your Agent API”) adapter.run(port=8000)

run(host: str = '0.0.0.0', port: int = 8000, **kwargs)

Run the FastAPI server.

Parameters:

• host – Host to bind to

• port – Port to bind to

• **kwargs – Additional uvicorn arguments

class aw_agents.ToolExecutionResult(success: bool, data: Any = None, message: str | None = None, warnings: list[str] | None = None, metadata: Dict[str, Any] | None = None)

Standard result format for tool execution.

classmethod error_result(message: str, data: Any = None, **kwargs) → ToolExecutionResult

Create an error result.

classmethod success_result(data: Any, message: str | None = None, **kwargs) → ToolExecutionResult

Create a success result.

to_dict() → Dict[str, Any]

Convert to dictionary format.

aw_agents.create_api_server_script(agent_class, output_path: Path, *, default_port: int = 8000)

Generate an API server script for an agent.

Parameters:

• agent_class – Agent class to wrap

• output_path – Path to write the script

• default_port – Default port for the server

aw_agents.create_json_schema(properties: Dict[str, Dict[str, Any]], required: list[str] | None = None, **kwargs) → Dict[str, Any]

Helper to create JSON schema for tool parameters.

>>> schema = create_json_schema(
...     properties={
...         'url': {'type': 'string', 'description': 'URL to process'},
...         'context': {'type': 'string', 'description': 'Context info'}
...     },
...     required=['url']
... )
>>> schema['type']
'object'
>>> 'url' in schema['required']
True

aw_agents.create_mcp_server_script(agent_class, server_name: str, output_path: Path)

Generate an MCP server script for an agent.

Parameters:

• agent_class – Agent class to wrap

• server_name – Name for the MCP server

• output_path – Path to write the script