Documentation

Getting started

An MCP (Model Context Protocol) server is a lightweight service that gives AI assistants like Claude access to specialised tools and data. Instead of training an AI on your domain data, you expose tools the AI can call at runtime.

Datatreehaus provides hosted MCP servers specifically built for AEC workflows — classification lookup, BIM data access, and more. Each server is accessed via a URL and an API key.

Quick start

  1. 01

    Sign up for a free account

    Create an account at datatreehaus.com/auth/signup. Your first API key is generated automatically.

  2. 02

    Add the server to Claude Desktop

    Claude Desktop requires Node.js to be installed. Then add the following to your claude_desktop_config.json:

    macOS: ~/Library/Application Support/Claude/claude_desktop_config.json

    {
      "mcpServers": {
        "datatreehaus": {
          "command": "npx",
          "args": [
            "-y",
            "mcp-remote",
            "https://classify.datatreehaus.com/mcp",
            "--header",
            "Authorization:Bearer dtk_live_YOUR_KEY_HERE"
          ]
        }
      }
    }

    Replace dtk_live_YOUR_KEY_HERE with your key from the dashboard. The example above connects the Uniclass classifier — swap the URL for any tool from the list below.

  3. 03

    Restart and use

    Fully quit and reopen Claude Desktop (Cmd+Q on Mac). On first start, mcp-remote downloads automatically — this takes a few seconds. Tools are loaded on demand to save context — Claude will call them automatically when relevant, or you can ask directly.

Available tools

Timber Joist & Rafter Advisor

Returns the minimum compliant solid softwood section for a given span, use type, and member spacing. Covers floor joists, ceiling joists, flat roof joists, and rafters for UK domestic construction. Derived from TDUK Knowledge Library span tables (C16 and C24 grades).

Free
Uniclass 2015

Keyword search and classification selection for Autodesk Revit AI assistants. Returns Uniclass 2015 codes and descriptions matched to natural language input.

Free

Authentication

All Datatreehaus MCP servers require an API key passed as a Bearer token:

Authorization: Bearer dtk_live_YOUR_KEY_HERE

When using the mcp-remote config above, this is handled automatically — just paste your key into the config once.

Keys are in the format dtk_live_ followed by 32 hex characters. Generate and manage keys from your dashboard.

Do not share your key publicly or commit it to source control. If a key is compromised, revoke it from the dashboard and generate a new one.

Rate limits

The free tier allows 500 requests per month per API key. When the limit is reached, requests return 429 Too Many Requests with a Retry-After header indicating when the limit resets (first of the following month).

Monitor your usage from the dashboard.

Troubleshooting

No tools appear in Claude Desktop

Tools are deferred by default — Claude loads them on demand rather than at startup. There is no longer a tool icon in the UI. Just ask Claude to use a tool by name, or ask a question that would naturally require it.

Server fails to load on first start

The first launch downloads mcp-remote via npx, which can take 5–10 seconds. If it fails, ensure Node.js is installed and npx is on your PATH.

Using nvm on macOS

Claude Desktop does not inherit your shell PATH. If Node is installed via nvm, replace "command": "npx" with the full path, e.g. "command": "/Users/you/.nvm/versions/node/v20.20.2/bin/npx". Find your path with which npx in Terminal.

401 Unauthorised

Your API key is missing or incorrect. Check the key in your config matches one from the dashboard, and that there is no space between Authorization: and Bearer in the header argument.

Changelog

2026-04

Timber Joist & Rafter Advisor beta

Preliminary section sizing for UK domestic timber members. C16 and C24 grades, four member types, BS EN 336 preferred finished sizes.

2026-04

Structural Slab Sizer beta

Preliminary depth and tonnage estimates for 12 concrete slab system types using trained LightGBM models.

2026-04

Uniclass 2015 v1.0.0

Initial release. Semantic lookup across Ss and Pr tables with configurable depth. 11,167 classification entries.