API (Application Programming Interface)

How APIs work

APIs follow a request-response model. A client application sends a request to an API endpoint (a specific URL), and the server processes the request and returns a response—typically in JSON or XML format.

A typical API call:

1. Client sends request: GET https://api.example.com/users/123
2. Server processes: Looks up user 123 in the database
3. Server returns response: {"id": 123, "name": "Alice", "email": "alice@example.com"}

Modern web APIs are built on HTTP/HTTPS, making them accessible to applications written in any programming language [1].

Types of APIs

REST (Representational State Transfer)

The most widely used API architecture. REST uses standard HTTP methods (GET, POST, PUT, DELETE) to operate on resources identified by URLs. Introduced by Roy Fielding in his 2000 doctoral dissertation, REST is an architectural style—not a protocol—built on constraints including statelessness and uniform interfaces [2].

Example: GET /products/456 returns product 456’s data.

SOAP (Simple Object Access Protocol)

A strict, XML-based messaging protocol with formal contracts (WSDL) and built-in standards for security and transactions. Common in enterprise systems (banking, healthcare, government) requiring high reliability [1].

GraphQL

A query language developed by Facebook (2015) that lets clients request exactly the data they need through a single endpoint. Unlike REST’s fixed resource shapes, GraphQL returns client-specified fields—solving the over-fetching and under-fetching problems [3].

Webhooks

An event-driven “reverse API” where the server pushes data to a pre-registered URL when something happens (e.g., payment completed, order shipped). Webhooks eliminate the need for constant polling [4].

API categories by access level

• Open (Public) APIs: Available to any developer (e.g., OpenAI API, Google Maps API)
• Partner APIs: Shared with specific business partners under agreement
• Internal (Private) APIs: Used within an organization for system integration
• Composite APIs: Combine multiple calls into a single request

The API economy

APIs are the foundation of modern software. The global API economy was valued at approximately $17 billion in 2025 and is projected to reach $39 billion by 2030, growing at 17.7% CAGR [5]. Every cloud service, mobile app, and SaaS product relies on APIs for data exchange.

Why APIs matter for AI agents

APIs are how AI agents interact with the real world. When an agent books a flight, queries a database, or sends an email, it does so through API calls [6].

Function calling (OpenAI, 2023) allows LLMs to output structured JSON that maps to API endpoints, letting models invoke external tools [6].

Model Context Protocol (MCP) (Anthropic, November 2024) standardizes how AI systems discover and invoke tools exposed by MCP servers—decoupling tool definitions from the model layer. Adopted by OpenAI (March 2025) and Google DeepMind (April 2025), MCP is becoming the de facto standard for agent-tool integration [7].

Agent-to-Agent (A2A) (Google, 2025) addresses horizontal collaboration between specialized agents, complementing MCP’s vertical model-to-tool focus [8].

The API is the execution layer of the agentic economy: ACP, UCP, x402, and MPP all define how agents discover, negotiate, and pay for services through standardized interfaces.

Limitations

• Security risks: APIs expand attack surfaces—authentication, authorization, and rate limiting are essential
• Versioning challenges: Breaking changes can disrupt dependent systems
• Rate limits: Providers cap request frequency, which can constrain high-volume agent workflows
• Latency: Network round-trips add time, especially for chained API calls
• Documentation quality: Poorly documented APIs increase integration time and error rates

Reader questions

Q: Is an API the same as a website?
No. Websites are for humans to read; APIs are for software to communicate. A website returns HTML for browsers; an API returns structured data (usually JSON) for programs.

Q: Do I need to know programming to use an API?
Traditionally yes, but low-code tools and AI assistants are making APIs accessible to non-developers. Some APIs offer interactive consoles where you can test calls without writing code.

Q: What’s the difference between an API and an SDK?
An API is the interface (the contract). An SDK (Software Development Kit) is a library that makes calling that API easier in a specific programming language. SDKs wrap API calls in friendly functions.

Q: Why do AI agents need APIs?
Agents need APIs to take actions in the real world—sending messages, querying data, making payments, controlling devices. Without APIs, agents could only generate text but couldn’t do anything.