Third-Party Services: API and MCP — What They Are and Why to Connect Them

ASCN Agent's built-in integrations cover the most popular tools: Gmail, Google Calendar, Slack, Notion, GitHub. But every business has its own ecosystem. Some work in Airtable, some in HubSpot, some in Shopify. To let your agent work with any of these services, you need to understand two concepts: API and MCP.

What is an API

API (Application Programming Interface) is the way one program talks to another.

Think of a restaurant. You're sitting at a table and want to order a dish. You don't walk into the kitchen yourself — you tell the waiter. The waiter passes your order to the cooks, then brings your finished dish back to you.

In this analogy:

• You are the agent that needs information or an action
• The waiter is the API
• The kitchen is the service (Airtable, Stripe, Shopify, etc.)

Almost every modern web service has an API. Airtable, HubSpot, Stripe, Shopify, any CRM, any SaaS product — they all communicate through APIs.

For an agent to use an API, two things are needed:

• Documentation — a description of what the service can do and how to ask it
• An API key — something like a pass that confirms: "yes, this agent is authorized to access our data"

What is MCP

MCP (Model Context Protocol) is a modern standard built specifically for AI agents.

If an API is a common language between programs, then MCP is a language between programs and AI agents. The difference lies in how easily the agent can understand the tool.

Here's another analogy. Imagine you're hiring a new employee.

• API — you hand them a 500-page instruction manual and say: "figure it out yourself."
• MCP — a specialist shows up and tells you upfront: "here's what I can do, here's how to ask me, here's what I'll return."

When you connect an MCP server to an agent, the agent discovers its capabilities on its own. No need to explain commands manually — the agent understands the tool automatically.

The practical difference

Where to find MCP servers

MCP servers are published by the service developers themselves or by the community.

You can find community and official MCP servers in these directories:

• Smithery.ai — a catalog of ready-made MCP servers
• mcp.so — another catalog with filters
• GitHub — many services publish official MCP servers. Search: MCP server [service name]

You should be careful with catalog solutions — both MCP servers and ready-made skills for your agents. There is always a risk that a malicious file could be bundled inside. This puts your API keys at risk, so if you're using an agent for trading or handling private customer data, avoid unofficial APIs and MCP servers to eliminate that risk.

The safest approach is to use official MCP servers: just search for "company name + mcp" or ask your agent to find it. Companies typically link to their MCP from the official website, or it will be in the project's own documentation.

MCP is already available for Notion, GitHub, Slack, Google Drive, Airtable, Linear, Jira, and dozens of other tools.

1. Search for "Notion mcp" in your search engine
2. You'll see a page on the official domain https://developers.notion.com/guides/mcp/overview
3. Tell your agent "connect this MCP https://developers.notion.com/guides/mcp/overview" and the agent will handle the rest — you may only need to provide your account token.

How the agent uses all of this

If you're connecting via MCP:

1. Get the MCP server config (usually a JSON with a URL and settings)
2. Paste it into the agent settings: Brain → integrations/tools
3. Write to the agent in plain language: "check for new tasks in Linear"
4. The agent figures out what the tool can do and completes the task

If you're connecting via API:

1. Open the service's documentation (usually developer.service.com or docs.service.com/api)
2. Copy the relevant section of the documentation, or just the link
3. Get your API key from your account settings
4. Paste the documentation into the agent's knowledge base or directly into the chat
5. Add the API key to the agent's secrets
6. Give the agent a task — it figures out on its own how to call the right method

Usually all you need to provide is one thing: an API key. If the agent knows the documentation, it handles everything else itself.

Example

Carla runs an online store in Barcelona on Shopify. She wants the agent to check new orders every morning and send a summary to Telegram.

Shopify has an API. Carla opened the Shopify documentation, copied the section on retrieving orders, and added it to the agent's knowledge base. Then she added the API key to the secrets. She wrote to the agent: "Every day at 8:00, check new orders in Shopify and send a summary to my Telegram."

The agent understood the documentation and now sends the summary every morning without any reminders.