MCP connector vs custom GPT vs Claude project: which for a landlord?
A Claude project or a custom GPT with uploaded files is a good place for fixed material, like your lease template and house rules. It goes stale and holds copies of data. An MCP connector, or a custom GPT with an action pointed at an API, reads live records from your software. For questions about current rent, balances and leases, you want the live connection. Use files for rules and a connector for facts.
The three setups
| Setup | What it is | Data |
|---|---|---|
| Claude project | A workspace with its own chats, uploaded knowledge files and instructions | Copies you upload |
| Custom GPT | A ChatGPT assistant with instructions and knowledge files, and optionally actions that call an API | Files you upload, plus whatever the API returns |
| MCP connector | A link from Claude or ChatGPT to a server that exposes tools, using the Model Context Protocol | Live, read from the system behind it |
Anthropic describes projects as self-contained workspaces with their own chat histories and knowledge bases, and lets each have its own instructions. OpenAI describes GPT Actions as a way for ChatGPT users to interact with external applications via REST APIs, configured inside custom GPTs. The MCP site describes MCP as an open standard for connecting AI applications to outside systems.
When files are the right choice
- Your lease template, house rules, move-in checklist and standard letters. They change rarely.
- A style guide so every message sounds like you.
- Your own checklists, such as a turnover routine.
The risk is staleness. A rent roll you uploaded in March does not know a tenant paid in April. If you upload tenant data, you also have more copies of it to protect. A project is simple and works well for text that does not change.
When you want a live connection
- Who is late right now, and by how much.
- Which leases end in the next 90 days.
- This month's profit and loss.
- What is waiting for your approval.
A custom GPT with an action can do this if the software behind it publishes an API and someone has written the action. A connector can do it if the software offers an MCP server. Either way the live data stays in the system of record and the assistant fetches it when asked. Anthropic's privacy page also says raw content from connectors is not used for model improvement unless you copy it into a chat, which is another point in favor of reading rather than copying.
Questions to ask before you pick
- Can it change anything? Reads are low risk. Writes need a draft and your yes.
- Can it move money or sign something on its own? It should not.
- Does it read only your own portfolio?
- Is there a record of what it did?
- Can you disconnect it in one step?
- Who maintains it? A custom action you built breaks when the API changes.
You can combine the setups. A project with your lease template and house rules, plus a connector for live facts, is a common pairing.
How DoorKeeping handles it
DoorKeeping gives you an MCP connector for Claude and for ChatGPT, so there is nothing to write or maintain: add the address as a custom connector, connect, and choose the portfolio on the Allow screen. The tools are labeled as reads, drafts you say yes to, and items that ask for your Accept on doorkeeping.com. Your AI app can reject an item in your approvals queue but can never accept one. Rent and payments stay in your own bank accounts. DoorKeeping never holds your money and sends a payment only when you approve it. The steps and the full tool list are on the connect page.
Sources
- Anthropic: What are projects?, read Oct 30, 2026
- OpenAI: GPT Actions introduction, read Oct 30, 2026
- Model Context Protocol: What is MCP?, read Oct 30, 2026
- Anthropic: Is my data used for model training?, read Oct 30, 2026
- DoorKeeping: Connect Claude or ChatGPT (steps and tools list), read Oct 30, 2026
Questions
- Is an MCP connector the same as a custom GPT action?
- They solve a similar problem in different ways. A GPT action calls a REST API from a custom GPT. An MCP connector uses the open Model Context Protocol and works with apps that support it.
- Should I upload my rent roll to a Claude project?
- You can, but it will go stale and adds a copy of tenant data. A live connection avoids both.
- Do I need to code to use a connector?
- Not with DoorKeeping. You paste an address and connect.
- Can I use a project and a connector together?
- Yes. Keep fixed text like your lease template in the project and read live facts through the connector.