What is MCP Server?
MCP Server — A server that implements the Model Context Protocol, giving AI assistants access to external tools, databases, and APIs.
What is an MCP server?
An MCP server is a program that implements the Model Context Protocol (MCP) — an open standard that lets AI assistants connect to external tools and data sources. Instead of being limited to text generation, an AI with MCP access can query databases, run browser tests, call APIs, and interact with real systems.
How do MCP servers work?
MCP servers expose "tools" — specific capabilities that an AI assistant can invoke. When you configure an MCP server in your AI tool, the AI gains access to those tools and can use them as needed during conversations.
For example, the Playwright MCP server exposes browser automation tools. When an AI coding assistant has access to it, the AI can launch a browser, navigate to pages, click elements, and take screenshots — all to verify that code changes render correctly.
How do you set up an MCP server?
You add a JSON configuration block to your AI tool's config file:
{
"mcpServers": {
"playwright": {
"command": "npx",
"args": ["-y", "@playwright/mcp"]
}
}
}
Config file locations vary by tool: .mcp.json for Claude Code, .cursor/mcp.json for Cursor, ~/.codeium/windsurf/mcp_config.json for Windsurf.
MCP transport modes
MCP servers communicate using one of two transports:
| Transport | How It Works | Best For | |-----------|-------------|----------| | stdio | Spawned as a local process | Local tools, CLI integrations | | Streamable HTTP | Accessed via HTTP URL | Remote/hosted services |
What is the Agent Shelf MCP server?
Agent Shelf provides an MCP server that lets AI tools search, browse, and download agents directly from the registry. It's available as both a remote HTTP endpoint (https://www.agentshelf.dev/api/mcp) and a downloadable local package.