Model Context Protocol (MCP)
The Model Context Protocol (MCP) is an open standard, released by Anthropic in November 2024, that defines how applications supply context and tools to large language models. Before MCP, every integration between an AI application and an external system was a bespoke connector. MCP replaces those one-off integrations with a single protocol: a model host speaks MCP to any number of MCP servers, each of which exposes tools, resources, and prompts in a uniform shape. The protocol has become the common wiring for agentic AI, which also makes it a shared attack surface worth governing.
How the protocol is structured
MCP defines two roles: a host (the AI application, which embeds an MCP client) and a server (which exposes capabilities). They exchange JSON-RPC messages over a transport, either standard input/output for local servers or HTTP with server-sent events for remote ones. A server advertises three capability types: tools the model can invoke, resources it can read, and prompt templates it can fill. This uniform shape is what lets one agent compose capabilities from many servers without custom code for each.
Why the protocol needs a policy layer
A standard interface for tool use is also a standard interface for tool abuse. An instruction hidden in a document, an email, or even a tool's own description can direct an agent to call a capability the requester was never authorized to use. For HTTP-transport MCP traffic, DeepInspect reads the identity context of the calling agent, evaluates per-route and per-role policy before the call reaches the server, and writes a per-decision audit record. That record is the evidence trail showing which agent identity invoked which tool, and under which policy verdict.
Related reading
- Model Context Protocol Security: How the MCP Transport Layer Changes the Inspection Boundary
The Model Context Protocol standardizes how an LLM client connects to tool servers and exchanges context, tool calls, and tool results. The transport layer carries the agent identity, the tool call payloads, and the tool return values. The inspection boundary an MCP deployment owes is the HTTP leg between the MCP client and the MCP server. This piece walks through the transport modes MCP supports, the inspection target on each, the identity-aware policy decisions the deployment commits per call, and the audit record format that survives an Article 12 review.
- MCP Security Best Practices: Authorizing and Auditing Model Context Protocol Traffic
The Model Context Protocol gives an LLM a standard way to call tools, and over HTTP transport those calls are AI traffic you can authorize and audit. This covers the practices that matter at the request boundary: verify identity on every call, scope tools per role, treat tool descriptions as untrusted, fail closed, and log per decision. It also marks the line where local stdio servers fall outside a gateway.