AI MCP & Evals
Direct Model Context Protocol client, MCP agent loop, and evaluation suite
🔌 FastMCP Direct Client ⓘ
Direct FastMCP Client spawning cricket_server.py over stdio for deterministic JSON-RPC tool calls.
Handshake with the MCP server, list exposed tools, and invoke get_live_score or get_player_stats directly without an LLM.
🤖 FastMCP Tool Agent ⓘ
OpenAI tool-calling agent discovering and invoking FastMCP server tools dynamically over stdio.
Ask a cricket question; the agent converts MCP tool schemas to OpenAI function definitions, calls the matching tool, and synthesizes a grounded answer.
🧪 Automated Evaluation Suite ⓘ
Automated 5-case evaluation harness asserting both MCP tool routing and ground-truth answer tokens.
Runs 4 positive cricket queries and 1 negative math test case to verify tool selection accuracy and factual grounding.