Skip to main content
Model Context Protocol (MCP) is an open standard that allows AI applications to securely connect with external systems using a common protocol. Instead of every AI app needing a custom integration for every service, MCP provides one consistent way to discover and use capabilities. Think of MCP like USB-C for AI. Rather than building a different connector for every application, an AI client connects to an MCP server, asks what capabilities are available, and invokes them using a standardized format. This makes integrations simpler, more portable, and easier to maintain.

​
Why MCP exists

Before MCP, every AI application had its own way of integrating with external APIs. Developers often had to build and maintain separate integrations for each client. With MCP, an API or service can expose its capabilities once through an MCP server, and any compatible client can use them. Instead of building many integrations: 
API → Claude
API → Cursor
API → VS Code
API → Codex
API → Windsurf
You build one: 
API → MCP Server → Any MCP-compatible client

​
The main parts

​
Client

The AI application that connects to an MCP server. Examples include Claude Code, Cursor, Codex, VS Code, Windsurf, and other MCP-compatible applications. A client discovers available capabilities, decides which ones to use, and sends requests to the server.

​
Server

The service that exposes capabilities through the Model Context Protocol. It tells clients what tools, resources, and prompts are available, validates incoming requests, and communicates with the underlying systems. With 0mcp, you don’t need to build or host this server yourself. 0mcp hosts and manages it for you.

​
Tool

A tool performs an action. Tools are usually backed by API operations and can retrieve or modify data. Examples:
  • Get an order
  • Search customers
  • Create a support ticket
  • Send an email
  • Update inventory
When an AI client needs to perform an action, it calls a tool.

​
Resource

A resource is information that an AI client can read. Unlike tools, resources don’t perform actions. They simply provide content. Examples:
  • Product documentation
  • Support policies
  • Pricing information
  • Internal knowledge
  • Markdown or JSON files
Resources help AI models answer questions with reliable, up-to-date information.

​
Prompt

A prompt is a reusable instruction template that can include optional inputs. Instead of rewriting the same instructions repeatedly, clients can reuse predefined prompts. Examples:
  • Summarize customer feedback
  • Generate release notes
  • Draft a support response
  • Analyze API logs
Prompts make AI workflows more consistent across teams and applications.
Use a tool for an action, a resource for read-only information, and a prompt for reusable instructions.

​
How clients use an MCP server

A typical request follows this flow:
  1. The client connects to an MCP server.
  2. The server advertises its available tools, resources, and prompts.
  3. The AI model decides which capability it needs.
  4. The client invokes that capability.
  5. The server executes the request and returns the result.
  6. The AI uses the result to continue the conversation.
This standardized workflow allows the same MCP server to work across many different AI clients.

​
How 0mcp fits in

0mcp turns your existing OpenAPI specification into a hosted MCP server. Instead of manually implementing the MCP protocol, you:
  • Import your OpenAPI document
  • Choose which API operations become tools
  • Publish your server
  • Connect any supported MCP client
0mcp handles hosting, protocol compatibility, versioning, and client connectivity so you can focus on your API.

​
Transport

0mcp uses Streamable HTTP, the modern transport defined by the MCP specification. Clients connect to your hosted MCP endpoint over HTTPS, eliminating the need to run a local process or maintain long-lived local connections. This makes deployment simpler while remaining compatible with modern MCP clients.