Local Model Suitability MCP

AI-powered evaluation of whether your local model is actually good enough for the task at hand.

The Problem

When you have both a local model (Ollama, LM Studio, etc.) and cloud APIs available, agents face a decision they cannot make intelligently alone:

Should I run this locally or send it to the cloud?

Getting this wrong in either direction is expensive:

• Wrong direction 1 — cloud when local works: You pay Claude Opus rates for a task a 7B model handles perfectly. At scale, this is thousands of dollars wasted monthly.
• Wrong direction 2 — local when cloud is needed: You run a complex reasoning task through a small model and get silent quality failures. The agent proceeds confidently on bad output.
• Wrong direction 3 — cloud when data is sensitive: You send confidential internal data to an external API that logs it. A privacy or compliance violation you never intended.

The Solution

evaluate_local_model_suitability is a single AI-powered tool that reasons across four dimensions simultaneously — cost, privacy, latency, and quality — and returns a clear verdict your agent can act on.

Verdict: LOCAL | CLOUD | EITHER | NEITHER

This is not a benchmark lookup. Claude reasons about your specific task, your specific model, and your specific constraints.

Installation

npx local-model-suitability-mcp

Or install globally:

npm install -g local-model-suitability-mcp

Claude Desktop / Claude Code config

{
  "mcpServers": {
    "local-model-suitability": {
      "command": "npx",
      "args": ["-y", "local-model-suitability-mcp"],
      "env": {
        "ANTHROPIC_API_KEY": "your-key-here"
      }
    }
  }
}

With Pro API key

{
  "mcpServers": {
    "local-model-suitability": {
      "command": "npx",
      "args": ["-y", "local-model-suitability-mcp"],
      "env": {
        "ANTHROPIC_API_KEY": "your-anthropic-key",
        "LMS_API_KEY": "your-pro-key-from-kordagencies"
      }
    }
  }
}

Tool: evaluate_local_model_suitability

Parameters

Example Request

{
  "task_description": "Classify customer support emails into 5 categories: billing, technical, returns, complaints, general. Must be accurate enough for production routing — wrong classification means wrong team gets the ticket.",
  "local_model": "llama3.1:8b",
  "quality_threshold": "production",
  "use_case_type": "classification",
  "data_sensitivity": "internal",
  "latency_requirement": "moderate"
}

Example Response

{
  "verdict": "EITHER",
  "confidence": "HIGH",
  "summary": "Llama 3.1 8B can handle 5-category email classification at production quality if emails are clear — use local to protect customer data and save cost, with cloud fallback for ambiguous cases.",
  "model_evaluated": "llama3.1:8b",
  "model_profile": {
    "parameter_count": "8B",
    "tier": "small",
    "known_strengths": ["simple Q&A", "basic summarisation", "short classification", "data extraction"],
    "known_weaknesses": ["complex multi-step reasoning", "long-context coherence", "nuanced instruction following"]
  },
  "task_complexity": "SIMPLE",
  "reasoning": {
    "quality_

... [View full README on GitHub](https://github.com/OjasKord/local-model-suitability-mcp#readme)