Mental Model

MCPMCPUSB-C for AI — one standard plug so any agent can connect to any data source or tool.The Model Context Protocol: an open standard (Anthropic, 2024) for connecting AI systems to external tools and data without building a custom integration for each one.Full definitionClose separates an agentAgentA worker you delegate to: brief it, and it takes steps on its own. A chatbot answers; an agent acts.An LLM that runs tools in a loop toward a goal — it acts, checks the result, decides the next step, and repeats until done.Full definitionClose client from external capabilities through a discoverable protocol. The client connects, negotiates capabilities, and exposes permitted primitives to the model.

Five Contracts

1. Primitive — tools, resources, prompts, or another documented capability.
2. Transport — how client and server communicate and recover.
3. Authentication — how the server obtains and refreshes authority.
4. Approval — what the user or policy must authorize before invocation.
5. Result — how output is bounded, attributed, and returned to model context.

Product support must be compared at this level. For example, Copilot cloud agent documents MCP tools while excluding resources, prompts, and remote OAuth. That is a different contract from Copilot CLI even though both are called MCP support.

Trust Boundary

The model sees a tool description, not the downstream authorization system. A server must validate input and enforce identity, scope, tenancy, and rate limits independently. Treat server output as untrusted external data that may be stale, malformed, or adversarial.

MCP expands reach. It does not turn an agent runtime into a policy engine.