Cross-LLM MCP Server

import { LLMProvider, ModelInfo, ModelTag, CostPreference, CostTier, } from "./types.js"; // Model registry last updated timestamp export const MODEL_REGISTRY_TIMESTAMP = new Date().toISOString(); // Comprehensive model registry with tags and metadata export const MODEL_REGISTRY: ModelInfo[] = [ // DeepSeek models { name: "deepseek-r1", provider: "deepseek", tags: ["coding", "reasoning", "math"], costTier: "flagship", description: "Flagship reasoning model, excels at coding, mathematics, and complex reasoning", }, { name: "deepseek-r1-distill-qwen-1.5b", provider: "deepseek", tags: ["coding", "math"], costTier: "budget", description: "Lightweight distilled version for efficient coding tasks", }, { name: "deepseek-r1-distill-qwen-7b", provider: "deepseek", tags: ["coding", "math"], costTier: "standard", description: "Balanced performance for coding and math tasks", }, { name: "deepseek-r1-distill-qwen-14b", provider: "deepseek", tags: ["coding", "math", "reasoning"], costTier: "standard", description: "Enhanced reasoning and coding capabilities", }, { name: "deepseek-r1-distill-qwen-32b", provider: "deepseek", tags: ["coding", "math", "reasoning"], costTier: "flagship", description: "High-performance distilled model for advanced coding and reasoning", }, { name: "deepseek-r1-distill-llama-8b", provider: "deepseek", tags: ["coding", "business"], costTier: "standard", description: "Balanced model for coding and business applications", }, { name: "deepseek-r1-distill-llama-70b", provider: "deepseek", tags: ["coding", "reasoning", "business"], costTier: "flagship", description: "Large distilled model for extensive coding and reasoning tasks", }, { name: "deepseek-chat", provider: "deepseek", tags: ["general", "coding"], costTier: "standard", description: "General-purpose chat model with coding capabilities", }, { name: "deepseek-coder", provider: "deepseek", tags: ["coding"], costTier: "standard", description: "Specialized coding model", }, // OpenAI/ChatGPT models { name: "gpt-4o", provider: "chatgpt", tags: ["coding", "business", "creative", "reasoning"], costTier: "flagship", description: "Latest flagship model with multimodal capabilities", }, { name: "gpt-4o-mini", provider: "chatgpt", tags: ["general", "coding"], costTier: "budget", description: "Cost-effective general-purpose model", }, { name: "gpt-4-turbo", provider: "chatgpt", tags: ["coding", "business", "creative"], costTier: "flagship", description: "High-performance model for complex tasks", }, { name: "gpt-4", provider: "chatgpt", tags: ["coding", "business", "creative", "reasoning"], costTier: "flagship", description: "Flagship GPT-4 model", }, { name: "gpt-3.5-turbo", provider: "chatgpt", tags: ["general", "coding"], costTier: "budget", description: "Fast and cost-effective general-purpose model", }, { name: "o1-preview", provider: "chatgpt", tags: ["reasoning", "math", "coding"], costTier: "flagship", description: "Advanced reasoning model for complex problem-solving", }, { name: "o1-mini", provider: "chatgpt", tags: ["reasoning", "math"], costTier: "standard", description: "Efficient reasoning model", }, // Anthropic/Claude models { name: "claude-3.5-sonnet-20241022", provider: "claude", tags: ["coding", "business", "reasoning", "creative"], costTier: "flagship", description: "Latest Claude model with enhanced capabilities", }, { name: "claude-3-opus-20240229", provider: "claude", tags: ["business", "creative", "reasoning"], costTier: "flagship", description: "Most capable Claude model for complex tasks", }, { name: "claude-3-sonnet-20240229", provider: "claude", tags: ["business", "coding", "general"], costTier: "standard", description: "Balanced performance model", }, { name: "claude-3-haiku-20240307", provider: "claude", tags: ["general"], costTier: "budget", description: "Fast and cost-effective model", }, // Google Gemini models { name: "gemini-2.0-flash-exp", provider: "gemini", tags: ["general", "creative"], costTier: "standard", description: "Experimental fast model", }, { name: "gemini-1.5-pro", provider: "gemini", tags: ["business", "creative", "reasoning"], costTier: "flagship", description: "High-performance model for complex tasks", }, { name: "gemini-1.5-flash", provider: "gemini", tags: ["general", "coding"], costTier: "budget", description: "Fast and efficient general-purpose model", }, { name: "gemini-2.5-flash", provider: "gemini", tags: ["general", "coding"], costTier: "standard", description: "Latest fast model with improved capabilities", }, { name: "gemini-pro", provider: "gemini", tags: ["general", "business"], costTier: "standard", description: "General-purpose model", }, // Mistral models { name: "mistral-large-latest", provider: "mistral", tags: ["business", "coding", "reasoning"], costTier: "flagship", description: "Flagship Mistral model", }, { name: "mistral-medium-latest", provider: "mistral", tags: ["business", "general"], costTier: "standard", description: "Balanced performance model", }, { name: "mistral-small-latest", provider: "mistral", tags: ["general"], costTier: "budget", description: "Cost-effective model", }, { name: "pixtral-large-latest", provider: "mistral", tags: ["creative", "general"], costTier: "flagship", description: "Multimodal model for creative tasks", }, // Perplexity models { name: "sonar-pro", provider: "perplexity", tags: ["business", "reasoning"], costTier: "flagship", description: "Premium model with web search capabilities", }, { name: "sonar-medium-online", provider: "perplexity", tags: ["general", "business"], costTier: "standard", description: "Balanced model with web search", }, { name: "sonar-small-online", provider: "perplexity", tags: ["general"], costTier: "budget", description: "Cost-effective model with web search", }, // xAI/Grok models { name: "grok-beta", provider: "grok", tags: ["general", "creative"], costTier: "standard", description: "Beta model with real-time information", }, { name: "grok-2", provider: "grok", tags: ["general", "creative"], costTier: "standard", description: "Latest Grok model", }, { name: "grok-3", provider: "grok", tags: ["general", "creative"], costTier: "flagship", description: "Flagship Grok model", }, // Moonshot AI/Kimi models { name: "moonshot-v1-8k", provider: "kimi", tags: ["general", "coding"], costTier: "budget", description: "Cost-effective model with 8k context", }, { name: "moonshot-v1-32k", provider: "kimi", tags: ["general", "coding", "business"], costTier: "standard", description: "Model with 32k context window", }, { name: "moonshot-v1-128k", provider: "kimi", tags: ["general", "coding", "business"], costTier: "flagship", description: "Large context window model for extensive tasks", }, ]; // Helper functions export function getModelsByTag(tag: ModelTag): ModelInfo[] { return MODEL_REGISTRY.filter((model) => model.tags.includes(tag)); } export function getModelInfo( provider: LLMProvider, modelName: string, ): ModelInfo | undefined { return MODEL_REGISTRY.find( (model) => model.provider === provider && model.name === modelName, ); } export function getModelsByProvider(provider: LLMProvider): ModelInfo[] { return MODEL_REGISTRY.filter((model) => model.provider === provider); } export function findModelByName(modelName: string): ModelInfo | undefined { return MODEL_REGISTRY.find((model) => model.name === modelName); } export function selectModelByPreference( provider: LLMProvider, tag?: ModelTag, costPreference?: CostPreference, ): string | undefined { let candidates = getModelsByProvider(provider); // Filter by tag if provided if (tag) { candidates = candidates.filter((model) => model.tags.includes(tag)); } if (candidates.length === 0) { return undefined; } // If only one candidate, return it if (candidates.length === 1) { return candidates[0].name; } // Apply cost preference if (costPreference === "flagship") { // Prefer flagship, then standard, then budget const flagship = candidates.filter((m) => m.costTier === "flagship"); if (flagship.length > 0) { candidates = flagship; } else { const standard = candidates.filter((m) => m.costTier === "standard"); if (standard.length > 0) { candidates = standard; } } } else if (costPreference === "cheaper") { // Prefer budget, then standard, then flagship const budget = candidates.filter((m) => m.costTier === "budget"); if (budget.length > 0) { candidates = budget; } else { const standard = candidates.filter((m) => m.costTier === "standard"); if (standard.length > 0) { candidates = standard; } } } // If multiple candidates remain, pick randomly (or first one) // Per plan: "let the LLM client decide, otherwise just pick a random one of the best match" const randomIndex = Math.floor(Math.random() * candidates.length); return candidates[randomIndex].name; } export function getAllTags(): ModelTag[] { return ["coding", "business", "reasoning", "math", "creative", "general"]; } export function getAllModels(): ModelInfo[] { return MODEL_REGISTRY; } // Get the model registry timestamp export function getModelRegistryTimestamp(): string { return MODEL_REGISTRY_TIMESTAMP; }