# aidc2026ai-melon/aidc-ai-mcp [Health: Active]

**Category:** 📐 Architecture & Design  
**Repository:** https://github.com/aidc2026ai-melon/aidc-ai-mcp  
**GitHub Stars:** 3  
**Views:** 4  
**Installs:** 0  
**Upvotes:** 0  
**Directory Page:** https://allmcps.com/mcp/aidc2026ai-melon-aidc-ai-mcp

## Description
Deterministic AI data-center design engine exposed as MCP tools (design, validate, layout): rack count, design PUE, total MVA, liquid/air cooling split, CDU planning, cost & timeline. NVIDIA Rubin-era, 22.9 kV intake, Korea. Remote Streamable HTTP at https://aidc-ai.io/api/mcp; no key for the anonymous tier.

## Tools
Capabilities this server exposes over MCP:

- **design** — Size an AI data center: returns rack count, PUE, total MVA, liquid/air cooling split, CDU count, cost (KRW), and build timeline.
- **validate** — Check a design against electrical, cooling, layout, safety, and data rules; returns severity-classified findings and RFIs.
- **layout** — Generate a rack-plan grid (hall dimensions, row/column positions in mm) and a site-block layout.

## Claude Desktop Quick Installation
Remote MCP endpoint (confidence: high). Install path detected from listing signals. Add as a URL/SSE server in your client:

```json
"mcpServers": {
  "aidc-ai-mcp": {
    "url": "https://aidc-ai.io/api/mcp"
  }
}
```

## Documentation

## What aidc2026ai-melon/aidc-ai-mcp MCP server does

The aidc2026ai-melon/aidc-ai-mcp MCP server connects MCP clients to the AIDC-AI.IO Design Engine, a remote service for sizing and checking AI data centers. Its calculations target Hopper, Blackwell, and NVIDIA Vera Rubin configurations, with regional assumptions for Korea, including a 22.9 kV utility intake and Korean authority-having-jurisdiction and code considerations.

The service accepts an IT load, rack density, GPU generation, site area, region, and optional redundancy, cooling-mode, and PUE-target settings. A design response includes deployment-unit-adjusted rack counts, design PUE, total MVA, liquid- and air-cooling loads, CDU count, estimated cost in KRW, build duration, and warnings.

## How it works

Communication uses MCP Streamable HTTP at `https://aidc-ai.io/api/mcp`. The three exposed tools are `design`, `validate`, and `layout`. The repository is a connector and usage example; the calculation engine, reference catalogs, and server-side rules are not published in the repository.

`validate` checks a supplied design against electrical, cooling, layout, safety, and data rules. Its response groups findings by severity and can include requests for information, along with pass, warning, and failure counts. `layout` produces both a rack-plan grid and a site-block plan. Rack output includes hall dimensions, rows, columns, and rack positions in millimeters; the site plan uses percentage-based block coordinates.

## Setup and configuration

Add the remote endpoint to an MCP client configuration. The README specifically provides examples for Claude Desktop, Cursor, and other Streamable HTTP clients. No credential is needed for the anonymous tier. An optional bearer token using the `aidc_live_<32hex>` format can provide a higher rate tier; registered or partner keys are obtained from the service operator.

The anonymous limit documented for the REST projection is 10 requests per hour on `/api/agent/*`. The same service also exposes REST equivalents for the three tools at `/api/agent/design`, `/api/agent/validate`, and `/api/agent/layout`, which can be called with JSON over POST.

The repository includes a Docker path for running the connector locally over stdio. The container connects to the hosted AIDC-AI.IO service by default, so local execution does not make the proprietary calculation engine local or open source.

## Tools and capabilities

- `design`: sizes an AI data center from inputs such as IT load, rack density, GPU generation, site area, and region.
- `validate`: evaluates a design against engineering and data rules and returns severity-coded findings and RFIs.
- `layout`: creates rack positions, hall dimensions, and a site-block layout.
- Supports `hopper`, `blackwell`, and `rubin` GPU generation values.
- Supports `air`, `hybrid`, and `liquid` cooling modes, plus `n`, `n_plus_1`, and `2n` redundancy options.

## Limitations and notes

The aidc2026ai-melon/aidc-ai-mcp MCP server does not publish the underlying engine, rack library, authority/code matrix, fabric topology, or direct-to-chip cooling models. Results depend on the hosted engine and its current reference data. The documented regional scope is Korea-focused, and outputs such as cost are expressed in KRW.

The example response in the README is illustrative rather than a guaranteed result. Input limits and accepted values should be respected: for example, IT load must be above zero and no more than 1,000 MW, rack density must be above zero and no more than 500 kW, and site area must be above zero and no more than 1,000,000 square meters. This repository's connector and examples are MIT licensed, while the hosted engine and its server-side logic remain proprietary.

_Full upstream README: https://allmcps.com/mcp/aidc2026ai-melon-aidc-ai-mcp/readme_