EX MCP Server

""" Performance tracking based on registry telemetry. - Aggregates success rate, avg latency - Basic recommendations utility - Optional history persistence via session store (future) """ from __future__ import annotations from statistics import mean from typing import Any, Dict, List from src.providers.registry import ModelProviderRegistry as Registry def aggregate_telemetry() -> Dict[str, Dict[str, Any]]: tel = Registry.get_telemetry() or {} out: Dict[str, Dict[str, Any]] = {} for m, t in tel.items(): succ = int(t.get("success", 0)); fail = int(t.get("failure", 0)) lat = [int(x) for x in (t.get("latency_ms") or [])] out[m] = { "calls": succ + fail, "success_rate": (succ / max(1, succ + fail)), "avg_latency_ms": int(mean(lat)) if lat else None, "input_tokens": int(t.get("input_tokens", 0)), "output_tokens": int(t.get("output_tokens", 0)), } return out def recommend_for_category(category: str) -> List[str]: tel = aggregate_telemetry() # Prefer higher success rate, then lower latency ranked = sorted(tel.items(), key=lambda kv: (kv[1]["success_rate"], -1 * (kv[1]["avg_latency_ms"] or 10**9)), reverse=True) return [m for m, _ in ranked]