MetaModel MCP Server

An MCP (Model Context Protocol) server that lets AI assistants use MetaModel's formula engine for certified computation. Turn published calculators, pricing tools, and engineering models into AI-callable tools — no hallucinated math.

All tools are read-only. This server fetches published project schemas and runs stateless computations. It never writes, modifies, or stores any data.

Quick Start

Remote (hosted) — easiest

No install. Add the URL as a custom connector:

https://www.metamodel.app/api/mcp

• Claude.ai / Claude Desktop: Settings → Connectors → Add custom connector → paste the URL
• Claude Code: claude mcp add --transport http metamodel https://www.metamodel.app/api/mcp

Works on web and mobile; no Node required. The hosted endpoint is stateless, read-only, and rate-limited.

Claude Desktop

Add to your Claude Desktop config (~/Library/Application Support/Claude/claude_desktop_config.json on macOS):

{
  "mcpServers": {
    "metamodel": {
      "command": "npx",
      "args": ["-y", "metamodel-mcp"]
    }
  }
}

Claude Code

claude mcp add metamodel -- npx -y metamodel-mcp

Other MCP Clients

Any MCP-compatible client can connect via stdio:

npx -y metamodel-mcp

Tools

metamodel_list_projects

Browse available calculators and engineering models. Returns project names, descriptions, publish tokens, and model names.

• Annotations: readOnlyHint: true

metamodel_get_schema

Discover inputs and outputs for a specific project. Returns model names, input parameters (with types, defaults, validation rules), output fields, and formulas.

• Annotations: readOnlyHint: true

Parameter	Type	Required	Description
token	string	Yes	Project publish token (from metamodel_list_projects)

metamodel_compute

Run a computation with input values. Send inputs once — they're auto-routed to the correct model by property name. Outputs from all models are returned, including cross-model cascading.

• Annotations: readOnlyHint: true

Parameter	Type	Required	Description
token	string	Yes	Project publish token
model	string	No	Target a specific model. When omitted, returns all models (recommended).
inputs	object	No	Input values as key-value pairs. Auto-routed to the correct model. Omitted inputs use defaults.

Examples

Example 1: List Available Projects

Prompt: "What calculators are available on MetaModel?"

Tool call: metamodel_list_projects()

Response (truncated):

{
  "projects": [
    {
      "token": "989bc6ae-e4d7-4585-8908-bfd03358043e",
      "name": "Deck Builder",
      "description": "Interactive deck builder with code-compliant sizing. Uses LOOKUP tables for beams, frost depths, and railing requirements.",
      "models": ["deck", "platformModel", "beamsModel", "postsModel", "railingModel", "summaryModel"]
    },
    {
      "token": "fdc5fbef-9d4b-46c3-b72d-1950e03d4bf4",
      "name": "Building Engineer",
      "description": "Complete building engineering calculator. Enter room dimensions to auto-size HVAC, electrical, and lighting systems with full cost estimation.",
      "models": ["room", "hvac", "electrical", "lighting", "costEstimator"]
    },
    {
      "token": "363c7753-ac4b-4de4-b1e7-e4536cb2f9bb",
      "name": "Life Expectancy Calculator",
      "description": "Estimate your life expectancy based on age, sex, and lifestyle factors.",
      "models": ["basics", "lifestyle", "results"]
    }
  ]
}

Example 2: Get Project Schema

Prompt: "What inputs does the Building Engineer calculator need?"

Tool call: metamodel_get_schema({ token: "fdc5fbef-9d4b-46c3-b72d-1950e03d4bf4" })

Response (truncated):

{
  "projectName": "Building Engineer",
  "models": [
    {
      "name": "room",
      "title": "Room Dimensions",
      "inputs": [
        { "name": "length", "type": "number", "defaultValue": 35.26 },
        { "name": "width", "type": "number", "defaultValue": 60 },
        { "name": "height", "type": "number", "defaultValue": 12 }
      ],
      "outputs": [
        { "name": "floorArea", "formula": "floorArea = length * width" },
        { "name": "volume", "formula": "volume = length * width * height" }
      ]
    },
    {
      "name": "hvac",
      "title": "HVAC Sizing",
      "inputs": [
        { "name": "climateZone", "defaultValue": "cold" },
        { "name": "btuFactor", "type": "number", "defaultValue": 25 }
      ],
      "outputs": [
        { "name": "btuRequired", "formula": "btuRequired = volume@room * btuFactor" },
        { "name": "tonnage", "formula": "tonnage = btuRequired / 12000" },
        { "name": "annualCost", "formula": "annualCost = tonnage * 120" }
      ]
    }
  ]
}

Note the cross-model references: volume@room means "use the volume output from the room model." MetaModel handles this cascading automatically.

Example 3: Run a Computation

Prompt: "Size the HVAC, electrical, and lighting for a 50×80 ft room with 14 ft ceilings"

Tool call: metamodel_compute({ token: "fdc5fbef-...", inputs: { length: 50, width: 80, height: 14 } })

Response:

{
  "models": {
    "room": {
      "outputs": { "floorArea": 4000, "volume": 56000, "wallArea": 3640 }
    },
    "hvac": {
      "outputs": { "btuRequired": 1400000, "tonnage": 116.7, "annualCost": 14000 }
    },
    "electrical": {
      "outputs": { "outletsNeeded": 304, "totalAmps": 456, "circuitCount": 23 }
    },
    "lighting": {
      "outputs": { "totalLumens": 201240, "fixtureCount": 51, "totalWattage": 2040 }
    },
    "costEstimator": {
      "outputs": {
        "hvacMaterial": 415917, "electricalMaterial": 18400,
        "lightingMaterial": 7650, "totalLabor": 89321, "grandTotal": 531288
      }
    }
  },
  "metadata": { "projectName": "Building Engineer", "evaluationMs": 7.1 }
}

One API call → room geometry, HVAC sizing, electrical load, lighting design, and full cost estimate. All computed from real formulas in ~7ms, not LLM-generated.

Development

To test against a local MetaModel instance:

# Clone and build
cd packages/mcp-server
npm install
npm run build

# Run with local URL
node dist/index.js --url http://localhost:3000

# Test with MCP Inspector
npx @modelcontextprotocol/inspector node dist/index.js --url http://localhost:3000

Privacy Policy

MetaModel's MCP server is stateless and read-only:

• No authentication required — all published projects are public
• No personal data collected — computations are anonymous
• No data stored — inputs are evaluated and discarded immediately
• No cookies or tracking — the server makes direct API calls to metamodel.app
• No third-party data sharing — results go only to the requesting client

Full privacy policy: https://www.metamodel.app/privacy

About MetaModel

MetaModel lets you build calculators, pricing tools, and engineering models with a spreadsheet-like formula language. Describe what you want, AI builds it, publish in seconds. Models become API-callable computation endpoints that any AI assistant can use via this MCP server.

License

MIT

Troubleshooting

• "Project not found or not published" — the token belongs to an unpublished or deleted project. Ask the project owner for a current publish token, or list what's available with metamodel_list_projects.
• Unknown input names — inputs are matched by property name. Call metamodel_get_schema first; it returns the exact input names, types, and defaults for the project.
• Inputs sent as a JSON string — pass inputs as a JSON object ({"width": 12}), not a stringified object. Stringified inputs fail with an "Unknown input" error naming a character index.
• Rate limits (hosted endpoint) — the hosted URL is rate-limited per client. For heavy use, run the npm package locally with your own network egress.
• Node version (local install) — the stdio server needs Node 18+.