Model Context Protocol (MCP)

Definition

Model Context Protocol (MCP) is a protocol and set of conventions for structuring, assembling, and governing the context provided to a language model during inference, including data, tools, state, and policy constraints.

Purpose

The purpose of MCP is to make model behavior more reliable, scalable, and auditable by defining how contextual inputs are selected, normalized, ordered, and bounded before a model is invoked.

Key Characteristics

• Explicit segmentation of context components such as instructions, user input, retrieved data, tools, and memory
• Deterministic context assembly based on rules rather than ad-hoc prompt concatenation
• Source attribution and traceability for context elements
• Token budget management through prioritization, summarization, and pruning
• Policy enforcement for access control, compliance, and safety constraints at runtime

Usage in Practice

In practice, MCP is used in agentic and enterprise AI systems to integrate multiple data sources and tools into a consistent context payload, ensuring that each model call receives the most relevant information within the context window under applicable policies.

Common Misconceptions

• MCP is only a longer or better-written prompt
• MCP replaces retrieval-augmented generation (RAG) rather than governing how retrieval outputs are used
• MCP guarantees correct model outputs regardless of context quality or tool design

Related Terms