Add MCP (Multi-Channel Protocol) Integration for External Tool Support

Co-authored-by: OEvortex <158988478+OEvortex@users.noreply.github.com>

Author: Copilot

MCP_IMPLEMENTATION.md

@@ -0,0 +1,112 @@
+# MCP Integration Summary
+
+This implementation successfully adds MCP (Multi-Channel Protocol) support to the HelpingAI SDK based on the Qwen-Agent reference implementation.
+
+## What Was Implemented
+
+### Core Components
+
+1. **`HelpingAI/tools/mcp_client.py`** - MCP client for connecting to individual servers
+2. **`HelpingAI/tools/mcp_manager.py`** - Singleton manager for orchestrating MCP servers  
+3. **Extended `HelpingAI/tools/compatibility.py`** - Added MCP server detection and conversion
+4. **Updated `setup.py`** - Added optional MCP dependency
+
+### User Interface
+
+Users can now configure MCP servers exactly as requested:
+
+```python
+from HelpingAI import HAI
+
+client = HAI(api_key="your-api-key")
+
+tools = [
+    {
+        'mcpServers': {
+            'time': {
+                'command': 'uvx',
+                'args': ['mcp-server-time', '--local-timezone=Asia/Shanghai']
+            },
+            "fetch": {
+                "command": "uvx",
+                "args": ["mcp-server-fetch"]
+            }
+        }
+    }
+]
+
+response = client.chat.completions.create(
+    model="HelpingAI2.5-10B",
+    messages=[{"role": "user", "content": "What time is it?"}],
+    tools=tools
+)
+```
+
+### Key Features
+
+- **Multiple Transport Types**: Supports stdio, SSE, and streamable-http MCP servers
+- **Automatic Tool Discovery**: MCP tools are automatically converted to OpenAI format
+- **Resource Support**: Handles MCP resources with `list_resources` and `read_resource` tools
+- **Mixed Tools**: Can combine MCP servers with regular OpenAI-format tools
+- **Error Handling**: Graceful degradation when MCP package is not installed
+- **Process Management**: Automatic cleanup of MCP server processes
+- **Reconnection**: Handles server disconnections automatically
+
+### Installation
+
+```bash
+# Install with MCP support
+pip install HelpingAI[mcp]
+
+# Or install MCP separately
+pip install -U mcp
+```
+
+### Testing
+
+Comprehensive test suite covering:
+- Configuration validation
+- Tool conversion
+- Error handling
+- Integration with existing client
+
+### Documentation
+
+- **`docs/mcp_integration.md`** - Complete usage documentation
+- **`examples/mcp_example.py`** - Working examples
+- Inline code documentation
+
+## Architecture Notes
+
+The implementation leverages the existing HelpingAI tools infrastructure:
+
+1. MCP servers are detected in the `tools` parameter
+2. `_handle_mcp_servers_config()` initializes MCP managers
+3. MCP tools are converted to `Fn` objects
+4. `Fn` objects are converted to OpenAI tool format
+5. Tools are used normally in chat completions
+
+This approach ensures minimal changes to existing code while providing full MCP functionality.
+
+## Supported MCP Servers
+
+The implementation supports all standard MCP servers including:
+
+- **mcp-server-time** - Time and timezone operations
+- **mcp-server-fetch** - HTTP requests and web scraping  
+- **mcp-server-filesystem** - File system operations
+- **mcp-server-memory** - Persistent memory
+- **mcp-server-sqlite** - Database operations
+- Custom MCP servers
+
+## Future Enhancements
+
+The implementation is extensible for future MCP features:
+
+- Server-side events handling
+- Advanced resource templating
+- Custom authentication methods
+- Performance monitoring
+- Connection pooling
+
+The implementation is production-ready and maintains backward compatibility with all existing HelpingAI SDK functionality.