DEV Community

DatanestDigital
DatanestDigital

Posted on • Originally published at decisionmatrix-mcp.pages.dev

DecisionMatrix MCP: give your AI agent a transparent, deterministic decision engine

#ai #showdev #productivity #mcp

Ask an AI agent to pick between three vendors, or a database, or a job offer, and it will happily give you an answer. Ask it to weigh five options against six weighted criteria and it quietly falls apart: inconsistent weights, arithmetic that drifts, and no way to see how it got there. "Decision-making" is exactly the kind of multi-step scoring LLMs are bad at — and exactly the kind of thing you don't want a black box for.

So I built DecisionMatrix MCP — a deterministic Model Context Protocol server that turns "which option is best?" into a transparent, reproducible calculation. You give it options and weighted criteria plus a score matrix; it returns a scored, ranked, and explained result: the winner, the full ranking, per-criterion breakdowns, the method used, the weights applied, and a plain-language explanation. Every number runs through decimal.js (never floats), so identical inputs always produce identical output.

It's live, free to start, and takes ~30 seconds to add.

Add it to your agent

Remote server over Streamable HTTP — no install:

https://decisionmatrix-mcp.pages.dev/mcp
Enter fullscreen mode Exit fullscreen mode

Generic client (Cursor, etc.):

{
  "mcpServers": {
    "decisionmatrix": {
      "url": "https://decisionmatrix-mcp.pages.dev/mcp"
    }
  }
}
Enter fullscreen mode Exit fullscreen mode

Claude Desktop (via the mcp-remote bridge):

{
  "mcpServers": {
    "decisionmatrix": {
      "command": "npx",
      "args": ["-y", "mcp-remote", "https://decisionmatrix-mcp.pages.dev/mcp"]
    }
  }
}
Enter fullscreen mode Exit fullscreen mode

Listed in the official MCP Registry as io.github.inity13/decisionmatrix-mcp.

What it does

Six tools:

  • create_decision — the main one: rank options against weighted criteria, return the winner + full ranking + per-criterion breakdown + explanation
  • score_options — the normalized scored matrix and ranking, without the narrative
  • sensitivity_analysis — how robust is the winner? Sweeps each criterion's weight ±20% and tells you which criteria could flip the result, and at what weight
  • compare_two — head-to-head of two options with a per-criterion breakdown
  • list_methods / health_check

Three scoring methods: weighted_sum, weighted_product, and TOPSIS (distance to the ideal/anti-ideal solution). Criteria can be benefit (higher is better) or cost (lower is better).

What a call looks like 

// create_decision
{
  "options": ["Postgres", "MongoDB", "DynamoDB"],
  "criteria": [
    { "name": "cost",         "weight": 3, "direction": "cost" },
    { "name": "scalability",  "weight": 5 },
    { "name": "team_familiarity", "weight": 4 }
  ],
  "scores": {
    "Postgres":  { "cost": 2, "scalability": 7, "team_familiarity": 9 },
    "MongoDB":   { "cost": 3, "scalability": 8, "team_familiarity": 6 },
    "DynamoDB":  { "cost": 5, "scalability": 9, "team_familiarity": 4 }
  }
}
Enter fullscreen mode Exit fullscreen mode

You get back the winner, a ranked list with exact scores, a per-criterion breakdown showing where each option gained or lost, the weights used, and a sentence explaining why. Change a weight and the result changes predictably — and you can prove it with sensitivity_analysis.

Why deterministic matters

The whole point of offloading a decision to a tool is trust. DecisionMatrix is stateless (no database, no sessions) and byte-for-byte reproducible. The hosted endpoint is a Cloudflare Pages Function; the same engine also runs as a local stdio server you can self-host with a one-line Docker build. MIT licensed.

Pricing

  • Free — 15 calls/day, no key needed
  • Starter — $12/mo — 5,000 calls/day
  • Pro — $39/mo — 50,000 calls/day

When an agent hits the free limit, the tool returns a structured error with the checkout URL, so an autonomous agent can surface the paywall and the user is two clicks from a key. Prefer to self-host? It's open source with a Dockerfile — run it with unlimited calls and your own keys.

Links

If your agents make choices — vendor selection, architecture, prioritization, hiring — give it a try. I'm considering adding AHP (with a consistency ratio) and Pareto/efficiency-frontier tools next; tell me what you'd want.

Top comments (0)