Support for MCP

[migduroli]

This pull request introduces comprehensive support for the Model Context Protocol (MCP) over JSON-RPC in the Flama framework. It adds new modules, endpoints, routing, protocol error handling, and response types to enable extensible and standards-compliant machine learning model interaction. The changes also generalize JSON-RPC support across the framework, including new exception handling and endpoint base classes.

Major enhancements and new features:

Model Context Protocol (MCP) Integration

• Added the MCPModule, which manages one or more MCP servers, and provides decorators and methods to register tools, resources, and prompts for model interaction. (flama/mcp/module.py, flama/applications.py) [1] [2] [3] [4]
• Implemented the MCPServer and MCPRoute classes, allowing MCP servers to be mounted at specific paths and routed via HTTP POST requests. (flama/mcp/server.py, flama/mcp/routing.py)
• Created the MCPEndpoint, which defines the MCP API over JSON-RPC, including methods for initialization, tool/resource/prompt management, and protocol-compliant responses. (flama/mcp/endpoint.py)

JSON-RPC Support

• Introduced a generic JSONRPCEndpoint base class for handling JSON-RPC requests, including method dispatch, error handling, and response formatting. (flama/endpoints/jsonrpc.py, flama/endpoints/__init__.py) [1] [2]
• Added JSONRPCException for protocol-specific error reporting, and integrated it into the debug middleware for consistent error handling. (flama/exceptions.py, flama/debug/middleware.py) [1] [2] [3] [4]

JSON-RPC Response Types

• Defined JSONRPCResponse and JSONRPCErrorResponse classes for standard-compliant JSON-RPC responses, along with a JSONRPCStatus enum for protocol error codes. (flama/http.py) [1] [2] [3]

Other improvements:

• Updated the HTTP endpoint dispatch method to return a value, aligning it with JSON-RPC endpoint expectations. (flama/endpoints/http.py)
• Ensured all new MCP and JSON-RPC modules are imported and available within the package. (flama/mcp/__init__.py)

These changes lay the foundation for robust, extensible, and standards-based model serving and interaction within Flama, making it easier to build interoperable AI systems.