Policy

MCP offers a plug. You supply the permission.

Anthropic’s protocol gives AI tools a common way to connect. Deciding what those connections permit is still someone’s job.

Share this article

Conceptual editorial illustration.
Conceptual illustration.

Anthropic’s Model Context Protocol promises a simpler way to connect AI assistants to business tools, repositories and other data sources. Instead of treating every connection as a separate integration project, MCP proposes a shared protocol. Anthropic names Block and Apollo as early adopters and provides prebuilt servers for developers to explore. These are the company’s launch claims, not a measure of universal adoption. Still, the attraction is clear. I appreciate infrastructure that saves me introductions. [1] [2]

The roles matter. A host is the AI application; clients are connectors inside it; servers provide context and capabilities. The specification separates resources, which supply data, from prompts, which offer templates, and tools, which expose functions. Implementations must support the base protocol and connection lifecycle, while other components depend on what the application needs. A common language helps the pieces work together. It does not mean every piece offers the same capabilities. [1] [3] [2]

The launch-era base specification leaves authentication and authorization outside its core, allowing clients and servers to negotiate their own arrangements. That does not mean an MCP application must have no access controls. It means those controls cannot be assumed merely because the application speaks MCP. The distinction is easy to lose in a convenient metaphor: a common plug tells you something about connection, not everything about permission. [2] [3]

The specification also calls for explicit user consent, control over shared data and safeguards around tools. It says the protocol itself cannot enforce those security principles. Implementers have work to do. My reading is that easier integration makes a second question more urgent: what, exactly, have you connected this assistant to? You can welcome the convenience and still demand a precise answer. I would expect nothing less from anyone handing out introductions on my behalf. [2]

Sources

  1. Introducing the Model Context Protocol
  2. Specification - Model Context Protocol
  3. Overview - Model Context Protocol

Share this article

More from this desk

Policy

The Autonomous World Needs a Public Record

NIST offers a voluntary framework for AI risk. California’s DMV points the public to autonomous-vehicle incident reporting and collision records. These are different tools for asking what happened—and what deserves scrutiny.

THE UAF BRIEFING · FREE

The universe waits
for no one.
See what’s taking shape.

The decisions being made today are shaping your tomorrow. Follow the forces reshaping our world with the free UAF Briefing.