7
1 Comment

This new protocol lets AI use any API. Here's how

I’ve built and shipped software long enough to know this: connecting APIs always takes more time than it should.

MCP changes that — by letting AI handle the setup for you.

If you’re an indie dev, this is for you.

And if you’re building a product with an API, keep reading — we’ll also cover how to make your app work directly with tools like ChatGPT and Claude.

What is MCP (and why it matters)

MCP (Model Context Protocol) is an open standard that lets AI tools like Claude and ChatGPT connect with software on their own.

It tells the AI:

  • What the tool can do
  • What info it needs
  • How to log in
  • What the outputs look like
  • And how to perform each action

You don’t need SDKs, wrappers, or custom code.

You just connect a tool and describe what you want done — the AI takes care of the rest.

Why this changes everything for indie devs

Say you want to collect form responses, save them to Notion, and get a short summary by email every Friday.

That’s useful — but it’s simple. You could build that with a Zap.

Now let’s look at what MCP can really do.

The old way:

  • You’d get new feedback from Jotform.
  • You’d get your customer list from Stripe.
  • You’d find the people who canceled and see what they said in their feedback.
  • You’d group their reasons (things like bugs, price, or missing features).
  • Then you’d make a Notion report with a short summary or a simple chart.
  • Post the key points in Slack every Friday.
  • Set up a script/cron to keep it running.

Takes hours. Involves 4 APIs, auth, mapping, and lots of testing.

The new way (with MCP):

  • Open Claude
  • Connect:
https://mcp.jotform.com
https://mcp.stripe.com
https://mcp.notion.com
Connect a Gmail MCP server
Connect a Slack MCP server
  • Then say something like:
"Every Friday, check new Jotform responses.Group the feedback by topic.Create a Notion page with charts.Email me a short summary."

That’s it.

The AI reads what each tool can do, connects them through MCP, and runs the whole thing safely — within each tool’s own permissions.

What can you actually do with MCP?

Here are a few one-sentence workflows you can run — right from chat:

  • “Make a Notion page for today’s meeting notes.”
  • “Look in Stripe and tell me who canceled this week.”
  • “Check Intercom for feedback and write a summary in Notion.”
  • “Get new form responses and email me a report.”
  • “When I post on Ghost, write a Twitter thread and send a newsletter.”

One sentence. Multiple tools. The AI runs them through MCP, within each tool’s limits.

How to try MCP in 5 minutes

1. Open Claude or ChatGPT Pro: These are AI chat apps. You’ll need the paid version to use tools.

2. Go to the "Tools" or "Configure" section. Each tool is a little different:

  • In Claude, look for Tools
  • In ChatGPT, go to Custom GPTs → Configure tools

3. Paste in a link to the tool you want to use. For example:

4. Log in and give permission: The AI will ask you to connect your account. Click to approve.

5. Now just ask it to do something: Try something simple like, “Create a Notion page called ‘Meeting Notes’ with today’s date.”

The AI will understand what the tool can do — and do it for you. It’s like having a junior developer you can talk to.

What if you want to add MCP to your app?

If your product has an API, you can make it usable by Claude or GPT in a few simple steps.

Here’s how:

  1. Follow the MCP spec on GitHub – it explains the format and structure.
  2. Define your actions – things your API can do (for example: create_document, list_customers). These live in a small JSON file on your server.
  3. Describe the inputs and outputs – also in that JSON file, so the AI knows what data to send and expect.
  4. Add OAuth or API key support – the same way you already secure your API.
  5. Host it at something like https://mcp.yourapp.com.

That’s it.

You don’t need a plugin or special support for one company’s model.

Any LLM that speaks MCP — Claude, ChatGPT, or future models — will just work.

If you’re building a dev-facing tool and don’t support MCP yet, you’re making it harder for users to integrate and automate with your product. Get started!

on December 3, 2025
  1. 1

    always takes more time than I expect. thanks for this step-by-step!

Trending on Indie Hackers
I built a text-to-video AI in 30 days. User Avatar 67 comments What 300 Builders Taught Us at BTS About the Future of App Building User Avatar 52 comments I built something that helps founders turn user clicks into real change 🌱✨ User Avatar 50 comments From a personal problem to a $1K MRR SaaS tool User Avatar 46 comments How An Accident Turned Into A Product We’re Launching Today User Avatar 29 comments I built an Image-to-3D SaaS using Tencent's Hunyuan 3D AI User Avatar 25 comments