How AI assistants use a service

Updated 2026-11-13

An AI assistant can use a SpreadAPI service in three ways. This page explains the three ways and helps you pick the one that fits, depending on whose AI it is.

What an AI does with a service

A service is your workbook with named inputs and outputs. An AI assistant such as Claude or ChatGPT sends input values and gets the calculated outputs back, exactly like any other caller. The numbers come from your formulas, not from the AI. See Why answers are calculated.

Everything about AI for one service is on its AI (KI) tab, which has two pages:

Definition: the texts that tell an AI what the service does and what each value means. See AI → Definition.
Connect (Verbinden): the three ways an AI reaches the service.

The three ways

Open AI → Connect. The page says: "Three ways for an AI to use this service. Pick by who the AI belongs to."

Way	Whose AI	What it reaches	Needs a token
Your own AI (Deine eigene KI)	yours	every service of your account that you allow; it can also create new ones	no, the connection is the access
Someone else's AI: the MCP address of this service	a customer's or a colleague's	this one service only	only if the service is protected
Any agent that can open an address	anybody's	this one service only, over plain HTTP	only if the service is protected

1. Your own AI: a connection to your account

You connect Claude, ChatGPT or another AI to your account once. It then reaches every service you allow, without a token, and it can also build new services from a spreadsheet, test them, publish them and calculate with them, within the rights you give it.

Address: https://spreadapi.io/mcp
You manage connections on the AI agents (KI-Agenten) page. See The AI agents page.
Steps: Connect Claude, Connect ChatGPT.
What the AI can do: What a connected agent can do.

2. Someone else's AI: the MCP address of one service

Every published service has its own MCP address that offers only this one service. You give the address to a customer or colleague; they add it to their AI as a custom connector. Their AI can calculate with the service and nothing else.

Address: https://spreadapi.io/api/mcp/service/<service_id>
See The MCP address of one service.

3. Any agent that can open an address: the discovery address

The discovery address describes the service in words and JSON. An agent reads it with GET and calculates with GET or POST. It needs no MCP support, only the ability to open a web address.

Address: https://spreadapi.io/d/<service_id>
See The discovery address.

Which one to pick

You want your own assistant to answer questions with your services, or to build services for you: use an account connection (way 1).
You want to hand one calculation to other people's assistants: give them the service's MCP address (way 2).
The agent cannot add an MCP server, or you are wiring the service into a script or an automation tool: use the discovery address (way 3).

Before you connect anything

Publish the service. The MCP address and the discovery address of a draft do not answer. See Publish and republish.
Fill in AI → Definition. Without descriptions the AI sees only names.
Decide whether the service is public or protected by a token. See Tokens and access.

Limits

Calls made by an AI count like any other call of the service. See Call limits.
AI calls appear on the Usage (Nutzung) tab under the channel AI. See Usage overview.

For developers and agents

Account address (all allowed services, building and running): https://spreadapi.io/mcp, MCP over streamable HTTP, tools such as list_services, create_service, run_service.
Service MCP address (one service, calculation only): https://spreadapi.io/api/mcp/service/<service_id>, tools spreadapi_calc, spreadapi_batch, spreadapi_get_details.
Discovery address (one service, plain HTTP): https://spreadapi.io/d/<service_id>, actions calculate and batch.
All three serve the same description of a service: the texts from AI → Definition plus types, limits, allowed values and defaults from the parameters.

FAQ

How can an AI use my SpreadAPI service?

In three ways: your own AI connected to your account, the MCP address of one service for other people's AI, and the discovery address that any agent can open over HTTP. All three are listed on the service's AI → Connect page.

Which AI assistants work with SpreadAPI?

Claude and ChatGPT connect by signing in, and Claude Code with one command. Any other AI client that supports MCP servers can connect with a key, and an agent without MCP can use the discovery address.

What is the difference between the account address and the service MCP address?

The account address https://spreadapi.io/mcp is for your own AI: it reaches all services you allow and can build new ones. The service address https://spreadapi.io/api/mcp/service/<service_id> offers exactly one service for calculation and is meant for other people's AI.

Do I need MCP to let an AI use my service?

No. The discovery address https://spreadapi.io/d/<service_id> works for any agent that can open a web address, and the normal REST API works too.

Where do I find the AI settings of a service?

Open the service and select the AI (KI) tab. Definition holds the descriptions and Connect (Verbinden) holds the addresses and connections.

Does the service have to be published before an AI can use it?

Yes for the service's MCP address and the discovery address: they answer only once the service is published. An AI connected to your account can also work on drafts and test them if its connection may edit.

Do AI calls count towards my plan?

Yes. A calculation made by an AI on a published service counts like any other call. Testing a draft through an account connection is not counted.

As an agent, which address should I use for a single service?

If you support MCP, use https://spreadapi.io/api/mcp/service/<service_id> and call spreadapi_calc. Otherwise open https://spreadapi.io/d/<service_id> with GET and follow its quickStart and actions.