Model Context Protocol (MCP)
An open, vendor-neutral protocol — frequently called "**USB-C for AI**" — that standardizes how reasoning engines (LLMs and agentic runtimes) discover, invoke, and exchange context with tools and data sources. Uses **JSON-RPC 2.0** over multiple transports. The MCP architecture cleanly decouples the reasoning engine from the data systems via a three-entity model: - **MCP Host** — the runtime environment housing the LLM (e.g., Claude Desktop, agentic IDE, agent orchestrator). - **MCP Client** — the connector inside the host that negotiates the protocol handshake, retrieves context, and formats tool calls. - **MCP Server** — the standalone microservice that securely exposes specific tools, temporal memory, or S3 resources to clients.
Definition
An open, vendor-neutral protocol — frequently called "**USB-C for AI**" — that standardizes how reasoning engines (LLMs and agentic runtimes) discover, invoke, and exchange context with tools and data sources. Uses **JSON-RPC 2.0** over multiple transports. The MCP architecture cleanly decouples the reasoning engine from the data systems via a three-entity model: - **MCP Host** — the runtime environment housing the LLM (e.g., Claude Desktop, agentic IDE, agent orchestrator). - **MCP Client** — the connector inside the host that negotiates the protocol handshake, retrieves context, and formats tool calls. - **MCP Server** — the standalone microservice that securely exposes specific tools, temporal memory, or S3 resources to clients.
Pre-MCP, every agentic integration required bespoke API connectors — custom Boto3 logic, custom database adapters, custom file-read tools. MCP replaces that brittle per-integration glue with a uniform handshake that exposes capabilities at runtime, so an agent can dynamically discover what an MCP-attached server offers and call it through a standard JSON-RPC interface. The S3-relevance is direct: the AWS-published [MCP Server for AWS S3](https://aws.amazon.com/blogs/storage/implementing-conversational-ai-for-s3-tables-using-model-context-protocol-mcp/) lets an agent list buckets, read Iceberg tables via Daft, or append records to a data lake without the developer hardcoding any AWS SDK calls.
Standardized tool/resource exposure to LLMs, S3-backed agent memory accessed through MCP servers, Iceberg-table-as-MCP-resource for conversational analytics, agentic IDE integrations (Claude Desktop, Cursor, VS Code, JetBrains, Xcode), distributed multi-agent systems sharing a common discoverable operational framework.