What the UltraCart MCP server is
The UltraCart MCP server is a free, read-only connection between your AI assistant and the UltraCart documentation at docs.ultracart.com. MCP, the Model Context Protocol, is the standard way AI apps plug into outside tools. Once you add one web address, your assistant can search the merchant guides, read a full page, and look up any REST API call, instead of answering from memory.
That matters because AI tools are confident even when they are out of date. Pointing them at the real docs means a question like "how do I issue a partial refund?" gets the steps from the current guide, with the page it came from.
- Address:
https://mcp.ultracart.com - Cost and sign-in: free, no API key, no authentication
- Access: read-only documentation, nothing about your store and no way to change it
Connect your AI assistant
Every app asks for the same thing: the address above, the Streamable HTTP transport, and no authentication. We suggest the name ultracart-docs. The MCP setup page in the docs has the exact clicks and commands for each tool and stays current as those apps change.
| AI tool | How to add it |
|---|---|
| Claude (desktop and claude.ai) | Customize, then Connectors, then Add custom connector. Choose No sign in. |
| Claude Code | One command: claude mcp add --transport http --scope user ultracart-docs https://mcp.ultracart.com |
| ChatGPT | Add custom MCP server, paste the address, choose No authentication. |
| Codex | codex mcp add ultracart-docs --url https://mcp.ultracart.com |
| Cursor and VS Code | One-click install buttons on the setup page, or a short entry in your MCP settings file. |
| Gemini | gemini mcp add --transport http -s user ultracart-docs https://mcp.ultracart.com |
Some plans and company workspaces limit custom MCP servers, so an admin may need to allow it first. After you add it, ask a question about UltraCart and watch for the assistant to call one of the tools, which most apps show in the chat.
For merchants: ask in plain words
You do not need to know any tool names. Ask the way you would ask a colleague, and mention UltraCart or the docs so the assistant knows to look. These habits give the best answers:
- Name the task, not the feature. "Walk me through issuing a partial refund" beats "tell me about refunds."
- Say where you are. "I am editing an item and want to deliver a license key at checkout" lets the assistant find the right tab.
- Ask for the source. Add "cite the docs page" and open it to confirm the steps match your screen.
- Ask for options when there are several. Many UltraCart settings have a few modes. "What are my choices, and when is each one right?" pulls them all in.
- Keep your own data out. The server cannot see your store, so describe your setup in words and leave out customer details.
Here are three prompts that work well on day one:
- "Using the UltraCart docs, how do I set up free shipping for orders over a set amount? Cite the page."
- "What are the ways to deliver a license key automatically at checkout, and what does each one need?"
- "Explain the UltraCart data warehouse in plain terms and tell me what I can report on."
If a question goes beyond the docs, such as a pricing or account question, the assistant will not find an answer there. That is a good moment to contact the UltraCart team.
For developers: build with the API
This is where the server earns its keep. Instead of pasting API pages into a chat, you let the assistant find and read the exact operation, then write code against it. Five tools make that possible:
- search_docs: hybrid keyword and meaning search across the API reference, the developer guides, and the merchant guides. Narrow it with a scope of api, developer, merchant, or all.
- get_page: reads any docs page in full as Markdown.
- find_api_operation: finds a REST call by what it does, by request line, or by operation id. "Refund an order" returns
PUT /order/orders/{order_id}/refund. - get_api_operation: reads one operation in full: parameters, cautions, errors, and SDK setup. Pass
fieldto read a nested field such asorder.items, andlanguagefor a sample in cURL, JavaScript, TypeScript, Python, PHP, Java, C#, or Ruby. - get_ground_rules: the rules to follow before you write any code against the API.
A reliable workflow is to ask the assistant to read the ground rules first, find the operation, read it, and only then write the code. In your prompt, that sounds like: "Read the UltraCart ground rules, find the operation that lists orders, read it in Node, and write a script that fetches yesterday's orders with the official SDK."
Those ground rules cover the things that trip up first integrations:
- One request at a time. Parallel calls get a 429. The default budget is 1,000 requests per hour and 10,000 per day, per the rate limiting guide.
- A required version header. Every request needs
X-UltraCart-Api-Version. The official SDKs send it for you. - Expand what you need. Nested objects are left out unless you ask for them with
_expand. - Test on a developer account. UltraCart has no sandbox, so an API key acts on the account that issued it. Try anything that changes data on a few records first.
- Use BigQuery for bulk reads. The data warehouse has no REST rate limits, and the BigQuery SDK returns the same model objects.
Remember what is and is not connected. The assistant reads documentation through the server. It does not call your store's API for you.
You run the code it writes, with your own key, in your own environment. For the full API picture, see UltraCart API and webhooks and the developer quickstart.
What it does not do
Keeping expectations clear is how this stays useful. The server cannot see your orders, customers, or settings, and it cannot change anything in your account.
It also is not the same as the MCP servers setting on an UltraCart AI Agent. That setting lets your agent connect out to a tool server you choose, as covered in the AI Agent MCP servers guide. The UltraCart MCP server goes the other way: it lets your AI tool read our docs.
If you want your assistant to work with live store data, the closest options today are the data warehouse, which this BigQuery reporting guide covers, and the REST API with your own key.
What is next for StoreFronts
The docs server is the first piece of a bigger effort to make UltraCart work well with AI tools. We are building a StoreFront agent toolchain that lets an AI coding agent work on your StoreFront using the same elements you use, with nothing going live until you approve it. When your assistant already knows the docs, it makes better StoreFront changes. Watch this page and the newsletter for details.
UltraCart MCP server FAQ
Is the UltraCart MCP server free?
Yes. It needs no API key and no authentication, and it is read-only documentation access.
Does it connect to my store or change anything?
No. It only reads docs.ultracart.com. It cannot see your orders, customers, or settings, and it cannot make changes.
Which AI tools work with it?
Any app that supports MCP over Streamable HTTP. The setup page lists Claude, Claude Code, ChatGPT, Codex, Cursor, VS Code, and Gemini.
Do I need to be a developer?
No. Merchants can use it to get step-by-step answers from the guides. Developers get the API reference and code samples on top.
Is there another way to give an AI tool the docs?
Yes. The docs also publish llms.txt and a full llms-full.txt file, and most pages have a raw Markdown version if you add .md to the address.
Ready to try it? Add https://mcp.ultracart.com to your AI tool using the UltraCart MCP setup page, then ask it your next UltraCart question. If you would rather have a person map out an integration, UltraCart's professional services team can help.
