Reference
MCP hub
Understand MCP hosts, clients, servers, tools, resources, prompts, transports, and the security boundaries that come with them.
MCP expands what an AI app can reach. Treat every server as a permission boundary, not as a harmless plugin.
MCP in plain English
Model Context Protocol is a standard way for an AI app to connect to external systems through server-defined tools, resources, and prompts.
Host, client, server
The host is the AI app, the client is the protocol connector inside that app, and the server exposes capabilities such as file access, web fetches, repositories, or business systems.
Tools, resources, prompts
Tools let the AI ask a server to do work, resources expose readable context, and prompts provide reusable prompt templates or workflows.
Local and remote servers
Local servers usually run on your machine through stdio or local network transport. Remote servers run elsewhere and can introduce account, network, and token-passthrough risks.
stdio and Streamable HTTP
stdio usually launches a local process. Streamable HTTP uses network requests and needs careful origin, auth, localhost, and exposure review.
Security warnings
- File servers can read or write local data inside their configured scope.
- Developer servers may execute commands or access private repositories.
- API-key and OAuth servers can expose secrets if permissions are too broad.
- Remote servers can introduce SSRF, confused deputy, token passthrough, and unknown-permission risks.
- Localhost services can become network-exposed if binding, Docker ports, or firewall settings are wrong.
- Local AI Guide has not audited MCP server implementations.