Developers · Model Context Protocol

MCP server

Connect Proptonomy to ChatGPT, Claude, Cursor or your own assistant, and give it the same tools the in-app AI has for you — across every organization you work.

The server

URL: https://api.proptonomy.ai/mcp
Transport: Streamable HTTP — remote, so there is nothing to install or run locally
Sign-in: OAuth 2.1 with dynamic client registration and PKCE, which clients discover on their own, or a personal API key as a Bearer header
Discovery: https://api.proptonomy.ai/.well-known/mcp.json

Working several organizations

A key authenticates you, not one client. If you work Heimby and Avalcityrentals, one key reaches both — and an organization you are added to tomorrow works immediately, with no key to re-issue. Lose access to one and the connection loses it on the next call.

The tools themselves are scoped to one organization at a time, because every one of them (“search properties”, “create task”) has to act somewhere specific. Which one a connection operates in is decided by its URL:

/mcp: Your default organization.
/mcp/heimby: That organization by its slug — anything you can access.

So there are two ways to cover several clients, and they combine:

Add a connection per organization, using the same key, and get the full direct toolset in each. This is what most people want for day-to-day work.
Stay on one connection and name the organization when you ask: AskProptonomyAssistant takes an organization argument, so you can ask about Heimby and then Avalcityrentals in the same chat without adding anything.

ListProptonomyOrganizations returns everything the key reaches, with the exact URL for each — so the assistant can tell you what to add rather than you looking it up.

Naming an organization you cannot reach is an error, not a fallback

A request for an organization outside your access is refused by name. It never quietly falls back to a different client, because acting on the wrong one is far worse than failing.

ChatGPT

Custom connectors are available on ChatGPT Pro, Plus, Business, Enterprise and Edu. On a Business or Enterprise workspace an admin has to allow custom connectors first, under Workspace settings → Connected data.

1
Turn on developer mode
In ChatGPT on the web: Settings → Apps & connectors → Advanced settings → Developer mode. This is what lets a connector expose its full toolset rather than search-only.
2
Add the connector
Settings → Apps & connectors → Create, then paste the server URL. Choose OAuth as the authentication method.
```
https://api.proptonomy.ai/mcp
```
3
Sign in to Proptonomy
ChatGPT sends you to Proptonomy. Sign in, pick the organization, click Allow. Nothing to copy — ChatGPT registers itself with us automatically.
4
Use it in a chat
Open the + menu in the composer, enable proptonomy, and ask. ChatGPT shows you each tool call before it runs it.

Limit

ChatGPT’s deep research and company-knowledge features expect a connector to expose OpenAI’s search and fetch compatibility tools, which this server does not implement yet. Use it as a developer-mode connector in normal chats, where the full toolset is available.

claude.ai and Claude Desktop

1
Add a custom connector
Settings → Connectors → Add custom connector. Paste the server URL and save.
```
https://api.proptonomy.ai/mcp
```
2
Connect and sign in
Click Connect. You are sent to Proptonomy, sign in, pick the organization and click Allow. No key to copy.

Claude Code

Add the server once

claude mcp add --transport http --scope user proptonomy https://api.proptonomy.ai/mcp

2
Sign in
Start claude, run /mcp, choose proptonomy and sign in with your Proptonomy account.

Headless or CI? Use a personal key instead

Create one under Settings → AI clients & MCP in the app and add it as a header:

claude mcp add --transport http --scope user proptonomy https://api.proptonomy.ai/mcp --header "Authorization: Bearer <your key>"

Cursor, Windsurf and other MCP clients

Most clients read this block. Those that support OAuth sign in on their own; the rest take a personal key in the headers field.

{
  "mcpServers": {
    "proptonomy": {
      "type": "http",
      "url": "https://api.proptonomy.ai/mcp"
    }
  }
}

OpenAI API

Building your own assistant on GPT? The Responses API takes the server inline — no connector, no UI. Authenticate with a personal Proptonomy API key from Settings → AI clients & MCP.

{
  "model": "gpt-5",
  "tools": [
    {
      "type": "mcp",
      "server_label": "proptonomy",
      "server_url": "https://api.proptonomy.ai/mcp",
      "authorization": "<your Proptonomy API key>",
      "require_approval": "always"
    }
  ],
  "input": "Which of our arrivals tomorrow still have no cleaning task?"
}

What the assistant can do

Read and search properties, bookings and guests, the inbox, tasks, the knowledge base and SOPs, finance and reports.
Create and update tasks, knowledge entries, follow-ups, schedules and leads; send messages, emails and documents where your role allows it — tools that reach a person are marked destructive, so a well-behaved client confirms before running them.
Work with the connected PMS and operations tools (Guesty, Hostaway, Enso Connect, PriceLabs, Breezeway) when your organization has them.
Ask the Proptonomy assistant itself (AskProptonomyAssistant): the full agent loop with your organization’s context, continuing the same thread you see on the app’s Assistant page — and the one tool that can be pointed at any organization you work.

Every connection acts as you, inside one organization, with your permissions — nothing more. Review and revoke connections at any time under Settings → AI clients & MCP.