Last verified: November 2026.
A connector is a built-in link that lets your AI interact with another app. For example, with the connectors for Google’s apps, your AI can search your Google Drive files, edit Google Docs, and create Google Slides and Sheets.
If the software you want your AI to work in has no official connector, you can often still connect it. Ask your AI to research whether the software has one of three things: a third-party MCP server, an API, or a CLI. If it has any of them, your AI can usually build the connection for you.
This is more advanced than switching on an official connector, because you are setting up your own connector instead of using one the software company built and looks after. You don’t need to write any code, though. Your agent can read the software’s documentation, set up the connection, and test it. It will ask your permission along the way.
MCP servers, APIs and CLIs
There are three ways your AI can interact with another piece of software.
MCP server. This is the technical name for the connectors you see in your AI app. An MCP server is a piece of software that tells your AI how to interact with another app, with actions such as “find a contact” or “add a note to this volunteer’s record.” MCP stands for Model Context Protocol, the open standard they are built on. Some MCP servers are third-party: someone other than the software company built them, such as another business or a volunteer developer. They work the same way, but the company that makes the app did not make them and does not look after them.
API (application programming interface). APIs are the traditional way pieces of software talk to each other, and they existed long before AI. A software company’s API lets other programs read and change the data in its app. Many apps have one even when they have no AI connector. Your AI can write a small piece of code, or a skill, that uses it.
CLI (command-line interface). A CLI is very similar to an API, but it is made for people to use. You type commands into your computer’s terminal instead of clicking buttons. AI agents that run on your computer are good at using them too. Basecamp, for example, publishes an official command-line tool.
You don’t need to understand any of these in depth. Your AI knows what they are and can use them to build a connector for itself. All you need to do is ask your AI about them.
What you need
This works best in an AI agent that runs on your computer, such as Claude Code or OpenAI Codex. These agents can download tools, run code, and reach the software over the internet. They ask before they do.
A prompt to start with
I want to connect you to [NAME OF THE SOFTWARE] so that you can [WHAT YOU WANT IT TO DO, SUCH AS "READ OUR VOLUNTEER RECORDS AND DRAFT REMINDER EMAILS"]. It has no official connector. Please research whether it has:
1. A third-party MCP server
2. An API
3. A CLI (command-line tool)
For each one you find, tell me who made it, what it would let you read or change, what I would need to set up (such as an API key, a paid plan, or an administrator's approval), and any security concerns. Then recommend the simplest safe option and walk me through setting it up. Do not ask me to paste any API keys or passwords into this chat.
Once I have selected a method to connect you to this piece of software, can you then guide me through packaging up the connection as a callable skill that I can use?
API keys
Most APIs, and many CLIs and MCP servers, need an API key. An API key is a long code that works like a password for programs: anyone who has it can use the app as you, often with the power to change or delete your data. You usually create one in the software’s settings, under a heading such as “API,” “Developers,” or “Integrations.” Your AI can tell you where to look. Some apps only let administrators create keys, and some only offer API access on more expensive plans.
- Don’t share it with anyone. Keep it out of emails, shared documents, and team chats.
- Don’t paste it into your AI chat either. Ask your AI where to put it instead, then paste it there yourself.
- Give it only the access it needs. If the app lets you make a read-only key, start with one.
- Replace it if it leaks. If you think someone else has seen it, delete the key in the software’s settings and make a new one.
Before you rely on it
Before you connect software that holds supporter or volunteer records, check which data tier it falls into and whether your AI policy allows it. Anything your AI reads through the connection goes to the AI company.
Take the most care with third-party MCP servers. Anyone can publish one, and neither Anthropic nor OpenAI checks the ones you add yourself. Anthropic warns that a malicious server can contain hidden instructions that try to make Claude do things you did not ask for. Ask your AI who made the server, how widely it is used, and whether anyone still updates it. Only connect to servers you trust. If the software company offers its own API or CLI, using that means you only have to trust the company that already holds your data.
Test the connection before you build anything on it. Ask your AI to read back one record and check that it matches what you see in the app. Let it read data for a while before you let it change anything.
Your connector may break, for example if the software company changes its API. If it does, ask your AI to fix it for you.

