Garage Door Science

For developers & agent builders

What you can build with Garage Door Science.

Four distinct callers use the same eight tools in four different ways. None of them required us to build a bespoke integration. The OpenAPI spec is the contract; every caller consumes it.

Each tier below has a recipe in the cookbook repo with runnable code. Grab what you need.

Tier 1 · highest volume

Homeowner via general-purpose AI

Who: Jane, 42, garage door won’t close at 10pm. She opens ChatGPT or Claude on her phone, types what’s happening, and three exchanges later has a diagnosis and a local pro’s phone number. She never visits garagedoorscience.com.

The consumer AI she already has open is the entire UX layer. Our job is to make the tools callable: MCP for Claude, Custom GPT Actions for ChatGPT, REST for anything that speaks HTTP. The homeowner never sees us. For Claude Code users, there’s also an installable plugin — skill + MCP config bundled — at the claude-plugin repo.

Tools: diagnose → routeByZip → (optional) getInspectionReferencePhotos

{
  "mcpServers": {
    "garagedoorscience": {
      "command": "npx",
      "args": ["-y", "mcp-remote", "https://garagedoorscience.com/mcp"]
    }
  }
}
Recipe: Claude Desktop MCP setup

Tier 2 · higher per-caller value

Operators with their own AI tools

Who: Ari, ops lead at a property-management company with 3,000 rental units. Tenants report garage-door issues through the resident portal. His internal AI triages them: calls diagnose, classifies severity, auto-creates work orders or escalates.

Same eight tools. Different caller. Pro-tier bearer for higher rate limits, deeper retrieval. Insurance-claim adjusters, real-estate brokerages, and home-warranty companies fit the same pattern — structured answers at a cadence a human workflow can’t match.

Tools: diagnose + costEstimate + retrieveLabContext (pro tier, gds_live_ bearer)

curl -X POST https://garagedoorscience.com/api/v1/diagnose \
  -H "Authorization: Bearer gds_live_..." \
  -H "Content-Type: application/json" \
  -d '{"description":"grinding noise when door opens"}'

Tier 3 · highest leverage

Developers building on top of us

Who: Max is building a multi-tool home-services agent. Finds garagedoorscience.com via /brain. Wires our eight tools into his agent as a capability pack. One dev, one deploy — but thousands of homeowners reach our data through his agent indirectly.

For this caller, the abstraction that matters is the OpenAPI spec. LangChain, LlamaIndex, AutoGen, Postman, Scalar — they all consume schemas. Write the registry once, every framework imports it for free.

Tools: All 8 tools via OpenAPI spec import

from langchain_openai import ChatOpenAI
from langchain.agents import AgentExecutor, create_openai_tools_agent

# Import tools from the spec
# See langchain-agent/agent.py for the full example
llm = ChatOpenAI(model="gpt-4o")
agent = create_openai_tools_agent(llm, TOOLS, prompt)
AgentExecutor(agent=agent, tools=TOOLS).invoke({"input": query})
Recipe: LangChain agent recipe

Tier 3b · distinct product shape

Contractors embedding our tools in their site

Who: Jen owns a single-shop garage-door company. Wants her own website’s visitors to get smart diagnostic Q&A without hiring a developer. She pastes a 40-line HTML snippet into her site’s CMS and ships it the same day.

The API is callable from a browser directly. Vanilla fetch, no framework. The contractor’s site gains instant diagnostic capability branded to her shop, powered by our corpus and tools, with the lead handoff still flowing through our network. If she’s on WordPress, there’s a plugin that installs the widget via a [gds-diagnose] shortcode — no code-paste required.

Tools: diagnose + (optional) routeByZip, getInspectionReferencePhotos

<script>
fetch('https://garagedoorscience.com/api/v1/diagnose', {
  method: 'POST',
  headers: { 'Content-Type': 'application/json' },
  body: JSON.stringify({ description: userInput }),
}).then(r => r.json()).then(render);
</script>
Recipe: Contractor embed widget

Tier 4 · future / speculative

What's coming

Ambient voice-first agents on home hubs. Cross-vertical assistants once the white-label platform ships HVAC, plumbing, and solar. LLM training pipelines harvesting /llms.txt. None of these exist in their current form. All of them will consume the same tools if they use OpenAPI, which most do.

The pattern: build to the standard once. New callers arrive at no additional cost.

Getting started

  • Grab an API key at /developers (email magic-link, no credit card) — unlocks pro-tier rate limits and top-10 retrieval.
  • Browse the interactive docs at /developers/api (Scalar) to try tool calls from your browser.
  • Read the /brain page for a full architecture tour.
  • Fork the cookbook repo and ship.

Official packages

Beyond the cookbook (which is meant for copy-paste and customization), two installable packages ship the most common integrations as ready-to-use plugins:

  • wp-garagedoorscience — a WordPress plugin that adds a [gds-diagnose] shortcode to any page or post. Drop-in diagnostic widget for contractor and home-service sites; works on the public tier out of the box.
  • garagedoorscience-claude-plugin — a Claude Code plugin that ships a diagnose-garage-door skill plus the MCP connector config. Clone into ~/.claude/plugins/ and Claude gains eight garage-door tools.