Provides tools for discovering and comparing AI models hosted on Replicate, including pricing, performance benchmarks, and model information for community fine-tunes and custom pipelines.
Click on "Deploy Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type @ followed by the MCP server name and your instructions, e.g., "@AI Model Advisor MCP Server find cheapest Llama 3.3 70b"
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
The ultimate Model Context Protocol (MCP) server for AI model discovery, cost optimization, and performance benchmarking. Fully compatible with Claude Desktop, Cursor AI, Windsurf, and any MCP client.
Stop guessing which AI model to use. Give your agent the tools to compare pricing, intelligence benchmarks (MMLU/Coding), latency speed (TTFT), and throughput across 1000+ models from 5 top AI platform providers (OpenRouter, fal.ai, Together AI, Replicate, and Fireworks AI).
"Where is the cheapest Llama 3.3 endpoint?" "What is the fastest model for image generation under my budget?" "Which Claude 3.7 model has the highest coding benchmark?" Just ask your agent. It shops across all providers instantly with real-time AI model pricing.
New AI models drop constantly. Your AI coding agent (like Claude Desktop or Cursor) doesn't inherently know what's available, what API inference costs, or what the exact performance metrics are. This MCP fixes that by acting as a live model catalog and routing engine, giving your agent real-time access to:
š§ 300+ LLMs via OpenRouter API (GPT-4o, Claude 3.7, Gemini 2.5 Pro, Llama, Reasoning Models, DeepSeek V3/R1)
šØ 200+ Media Models via fal.ai API (Flux Pro, Stable Diffusion, Kling Video, Whisper)
ā” 200+ Open-Source Models via Together AI (Llama 3.3 70B, Qwen, Mistral)
š Community fine-tunes & LORAs via Replicate (Wan 2.1, Recraft, Custom pipelines)
š„ Blazing Fast Inference endpoints via Fireworks AI
š·ļø Cross-Provider Price Calculator ā find the cheapest API endpoint / lowest cost LLM for any architecture.
ā” Live Speed & Latency Metrics ā powered by Artificial Analysis (TTFT and Tokens/sec throughput).
š Intelligence Leaderboards ā baked-in MMLU, Math, and Coding benchmark scores.
š° Real-Time Pricing ā powered by our cloud backend (Zero config, no API keys needed for pricing).
ā Curated Quality Tiers (S/A/B/C) to prevent agents from picking hallucinating or outdated models.
š Discovery Engine ā agents can ask "what new AI models dropped this week?"
Related MCP server: LLM Benchmark MCP Server
The server works entirely out-of-the-box. Just add it to your MCP settings and save:
{
"mcpServers": {
"model-advisor": {
"command": "npx",
"args": ["-y", "ai-model-advisor-mcp@latest"]
}
}
}That's it! Live pricing for all 1,000+ models across all 5 providers is fetched automatically using our hosted Cloudflare Worker Pricing API. No API keys, no environment variables, no setup.
select_model_for_project ā NEWOne call for agents that already have project context. Returns the best overall model, cheapest acceptable model, and best value option without forcing the agent to manually chain several tools.
select_model_for_project({
project: "TypeScript MCP server for coding agents",
task: "coding assistant",
requirements: ["coding", "reasoning", "tool_use"],
expected_usage: { input_tokens: 5000000, output_tokens: 1000000 }
})The server searches the catalog behind the scenes and returns a compact decision with model IDs, provider, price, quality tier, reasons, tradeoffs, and structured output that agents can parse.
find_cheapest_provider ā NEWThe killer feature. Shop for a model across all 5 providers.
find_cheapest_provider({ model: "llama 3.3 70b" })Output:
š·ļø Price comparison for "llama 3.3 70b"
| Provider | Input $/1M | Output $/1M | Model ID |
|-------------|-----------|------------|--------------------------------------------------|
| OpenRouter | $0.00 | $0.00 | meta-llama/llama-3.3-70b-instruct |
| Together AI | $0.88 | $0.88 | meta-llama/Llama-3.3-70B-Instruct-Turbo |
| Fireworks | $0.90 | $0.90 | accounts/fireworks/models/llama-v3p3-70b-instruct |
š” Cheapest: OpenRouter ā FREEUses fuzzy matching ā handles version format differences (v3p3 = 3.3) across providers.
batch_get_pricing ā NEWGet pricing for multiple models in a single call. Returns a compact table.
batch_get_pricing({ model_ids: ["openai/gpt-4o", "anthropic/claude-sonnet-4", "fal-ai/flux-pro/v1.1", "meta-llama/Llama-3.3-70B-Instruct-Turbo"] })recommend_model"I need X" ā ranked models matching your task, requirements, and budget. Searches all 5 providers.
recommend_model({ task: "image generation", requirements: ["photorealistic", "fast"], budget: "low" })compare_modelsSide-by-side table across providers. Auto-adapts columns for LLMs vs media models.
compare_models({ model_ids: ["openai/gpt-4o", "anthropic/claude-sonnet-4", "google/gemini-2.5-pro-preview"] })list_modelsBrowse and filter by category, provider, capability, or price.
list_models({ category: "text-to-image", max_price: 0.05 })
list_models({ provider: "together", category: "llm" })
list_models({ provider: "replicate", category: "text-to-video" })get_model_infoComprehensive model card with everything you need to decide.
get_model_info({ model_id: "fal-ai/flux-pro/v1.1" })estimate_costCost estimation for any usage scenario.
estimate_cost({ model_id: "openai/gpt-4o", usage: { input_tokens: 1000000, output_tokens: 100000 } })
estimate_cost({ model_id: "fal-ai/flux-pro/v1.1", usage: { images: 500 } })whats_newDiscover recently added models. Never miss a new release.
whats_new({ since: "7d" })
whats_new({ since: "30d", category: "text-to-video" })Provider | Models | Type | Hosted Pricing Data |
350+ | LLMs | ā Yes | |
40+ | Image, Video, Audio, 3D | ā Yes | |
220+ | LLMs, Image | ā Yes | |
120+ | Everything | ā Yes | |
12+ | LLMs | ā Yes |
Category | Examples |
| GPT-4o, Claude, Gemini, Llama, Mistral, DeepSeek |
| Flux Pro, Stable Diffusion, DALL-E, Ideogram, Recraft |
| Flux Edit, img2img pipelines, upscalers |
| Kling, Minimax, Hunyuan, Wan 2.1 |
| Kling i2v, Runway, Wan i2v |
| Kokoro, ElevenLabs |
| Wizper (Whisper) |
| Music/sound generation |
| Hunyuan3D, Trellis |
| Visual understanding models |
| Text embedding models |
Popular models are rated on a curated quality scale:
S ā Best in class (Claude Sonnet 4, GPT-4o, Flux Pro Ultra, Kling v2.0)
A ā Excellent (Gemini Flash, Llama 3.3, Flux Dev, Recraft v3)
B ā Good (Mistral Small, Flux Schnell, SD3.5 Turbo)
C ā Adequate
Agent ā MCP Server ā Cloudflare Worker Pricing API (Our Hosted Backend)
āā Fetches from OpenRouter (350+ models)
āā Fetches from fal.ai (40+ models)
āā Fetches from Together AI (220+ models)
āā Fetches from Replicate (120+ models)
āā Fetches from Fireworks AI (12+ models)
Unified Model Registry
ā select, recommend, compare, list, info, estimate, shop, batchThe MCP Server connects to our blazing-fast Cloudflare Worker that aggregates live pricing data across all 5 providers on a recurring 6-hour cron schedule. This gives your agent real-time pricing awareness without requiring you to juggle 5 different API keys.
MIT
Get pricing for multiple models in a single call. Returns a compact table with pricing for all requested models. Much faster than calling get_model_info one at a time.
Inputs: model_ids (required).
Compare AI models side-by-side in a table. Works across providers and categories ā compare LLMs against each other, image generators, or even mix categories. Auto-adapts columns based on model types. Includes speed metrics (TTFT, tokens/sec) and benchmark scores (MMLU) when available.
Inputs: model_ids (required).
Estimate the cost of using an AI model. For LLMs, provide token counts. For media models, provide unit counts. Example: estimate_cost({model_id: "fal-ai/flux-pro", usage: {images: 100}})
Inputs: usage (required), model_id (required).
Find the cheapest provider for a given model. Searches across all 5 providers to find where a model (e.g. "llama 3.3 70b", "flux pro", "deepseek r1") is available and compares pricing. Uses fuzzy matching to find the same model across different providers.
Inputs: model (required).
Get a comprehensive model card for any AI model. Includes description, pricing breakdown, capabilities, quality tier, context length (LLMs), speed metrics (TTFT, throughput), benchmark scores (MMLU, coding), tags, license, and more.
Inputs: model_id (required).
List and filter available AI models from all 5 providers. Filter by category (llm, text-to-image, text-to-video, text-to-speech, speech-to-text, image-to-3d, etc.), provider (openrouter, fal, together, replicate, fireworks), capability, or max price.
Inputs: limit, category, provider, max_price, +1 more.
Recommend the best AI model for a task. Searches across 1000+ models spanning LLMs, image gen, video gen, TTS, STT, 3D, and more from 5 providers (OpenRouter, fal.ai, Together AI, Replicate, Fireworks). Returns ranked results based on task match, capabilities, quality tier, price, speed (TTFT/throughput), and intelligence benchmarks (MMLU/coding).
Inputs: task (required), limit, budget, requirements.
Pick the best overall, cheapest acceptable, and best value AI model for a project. Use this when an agent has project context and needs one decision instead of manually chaining recommend, compare, pricing, and cost tools. Returns concise markdown plus structured fields with candidates, scores, reasons, pricing, and tradeoffs.
Inputs: task, limit, budget, project, +3 more. Returns: query, best_value, candidates, cost_basis, +2 more
Discover recently added AI models. Answers "what new models dropped this week?" Filter by time window and category.
Inputs: limit, since, category.
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelogā¦
Scored across 9 tools
Most tools have clearly distinct purposes: listing, getting details, comparing, recommending, estimating cost, etc. However, recommend_model and select_model_for_project overlap in that both provide recommendations; select_model_for_project is explicitly a higher-level aggregation tool, but agents could be uncertain which to use.
The majority of tools follow a consistent verb_noun snake_case pattern (e.g., list_models, get_model_info, estimate_cost). The exception is whats_new, which deviates from the verb_noun structure but remains clear and fits the domain.
With 9 tools, the set is well-scoped for an AI model advisor. Each tool addresses a distinct need, and the count is within the ideal range for a focused server.
The tool surface covers the full lifecycle of model advisory: discovery (list_models, whats_new), detailed info (get_model_info), recommendation (recommend_model, select_model_for_project), comparison (compare_models), pricing (batch_get_pricing, find_cheapest_provider), and cost estimation (estimate_cost). No significant gaps are apparent.
Real-time, drift-verified LLM pricing across 31 providers with cost+quality model routing.
Sourced AI-model pricing and capability data ā compare and route to the cheapest capable model.
Compare, estimate, and deploy cloud infrastructure across AWS, GCP, and Azure for AI agents.
Compare LLM API prices, search models and providers, and access reviewed benchmark results.