- Introduced MCPHostCallbacks for optional host capabilities like sampling and logging.
- Implemented MCPHostLLMBridge to handle MCP Sampling requests and bridge to LLM service.
- Created models for structured data conversion between MCP SDK and internal data models, including tool content items, prompts, and resources.
- Enhanced error handling and logging for better traceability during sampling operations.
- Added a new `tooling` module to define a unified model for tool declarations, invocations, and execution results, facilitating compatibility between plugins, legacy actions, and MCP tools.
- Implemented `ToolProvider` interface for various tool providers including built-in tools, MCP tools, and plugin runtime tools.
- Enhanced `MCPManager` and `MCPConnection` to support unified tool invocation and execution results.
- Updated `ComponentRegistry` and related classes to accommodate the new tool specifications and descriptions.
- Refactored existing components to utilize the new tooling system, ensuring backward compatibility with legacy actions.
- Improved error handling and logging for tool invocations across different providers.
- Refactored model fetching logic to support various authentication methods for OpenAI-compatible APIs.
- Introduced new data models for LLM service requests and responses to standardize interactions across layers.
- Added an adapter base class for unified request execution across different providers.
- Implemented utility functions for building OpenAI-compatible client configurations and request overrides.
- Removed unused core action mirror functionality from PluginRunnerSupervisor.
- Simplified action and command execution logic in send_service.py.
- Introduced ComponentQueryService for unified component querying in plugin runtime.
- Enhanced message component handling with new binary component support.
- Improved message sequence construction and detection of outbound message flags.
- Updated methods for sending messages to streamline the process and improve readability.
