Open standard protocol by Anthropic enabling AI models to call external tools, read resources, and use prompt templates via JSON-RPC.
MCP Server
A process that exposes tools, resources, and prompts via the MCP protocol; the server you build in this course is a Node.js process.
MCP Client
The AI host that connects to MCP servers — e.g., Gemini Code Assist Agent Mode, Claude Desktop, or Cursor.
Tool
An MCP-exposed function the AI can invoke; defined with a name, description, and JSON Schema describing its input parameters.
Resource
An MCP-exposed data source (file, URL, DB query) the AI can read as context without executing code.
Prompt
An MCP-exposed reusable prompt template the AI client can invoke by name, optionally with arguments.
stdio Transport
MCP communication over standard input/output; used for local process servers launched by the AI client directly.
Streamable HTTP
Current remote transport for MCP (replaces the deprecated HTTP+SSE). Server-Sent Events transport for remote MCP servers over HTTP; used when deploying to Azure or a cloud host.
JSON Schema
Standard format describing the shape of a JSON object; MCP uses it to validate tool inputs before the AI calls a function.
Node.js
JavaScript runtime for building MCP servers in TypeScript/JavaScript; required for all servers built in this course.
SDK (@modelcontextprotocol/sdk)
Official npm package providing MCP Server and Client base classes; handles protocol negotiation automatically.
Tool Discovery
Process by which an MCP client calls tools/list to enumerate available tools from all connected servers.
Context Window
The AI model's working memory; MCP tools extend it by pulling in external data (DB results, file contents) on demand.
.mcp.json
Project-level config file (used by Claude Code, VS Code, and other hosts) listing MCP server processes to start automatically when the project is opened.
Tool Chaining
Using the output of one MCP tool as input to another in a multi-step AI workflow — e.g., BigQuery result feeds a GitHub issue.