← Atlas

Tools & standards

MCP: a common language for AI tools

MCP standardizes how applications discover tools and retrieve context. Its introduction, later foundation donation, and security boundaries are separate parts of the story.

Reviewed Sep 5, 20265 sourcesEvidence through Dec 9, 2025
Download Markdown
From the evidence

One interface across many integrations

Anthropic introduced the Model Context Protocol on November 25, 2024, with an open specification and implementations for connecting AI applications to external systems. In the November 2025 specification, tools expose callable operations with schemas, while resources expose addressable context. The host application decides how those capabilities enter its user experience.

From the evidence

Introduction was not the foundation donation

The governance milestone came on December 9, 2025. The Linux Foundation announced AAIF with MCP, Block’s goose, and AGENTS.md as founding contributions; Anthropic separately announced its MCP donation. A shared institutional home establishes stewardship, not proof that every client implements the same capabilities or that adoption claims have been independently audited.

Atlas interpretation

A protocol is not a permission decision

A common interface makes integrations reusable, but an application still has to choose which servers to trust and what an agent may do. Research tools should return evidence and dates clearly enough to inspect. Reading a curated corpus and authorizing an action on a private system are different responsibilities, even when both use the same protocol.

How the story developed

  1. MCP is introduced

    An open protocol gives tools and external context a shared integration surface.

    Open event record →
  2. A versioned interface defines tools and resources

    The specification separates callable operations from addressable context and supports structured tool results.

  3. MCP moves into a shared foundation

    Anthropic announces the donation to AAIF, co-founded with Block and OpenAI.

What remains open

  • Which capabilities does a particular client implement, and how does it show evidence?
  • How should applications distinguish untrusted source text from instructions?
  • Can a research result be reproduced after the underlying corpus changes?

Newer developments to investigate

Unreviewed research leads after this brief’s evidence cutoff.

Source notebook

What each source supports, and how to read it.

[1] Primary announcement · Anthropic · Nov 25, 2024

Introducing the Model Context Protocol ↗

The introduction date; does not document the later foundation donation.

[2] Versioned specification · Model Context Protocol · Nov 25, 2025

Tools — Model Context Protocol specification ↗

Schemas, structured results, and responsibilities at the tool boundary.

[3] Versioned specification · Model Context Protocol · Nov 25, 2025

Resources — Model Context Protocol specification ↗

Addressable context and the application’s role in exposing it.

[4] Primary announcement · Anthropic · Dec 9, 2025

Donating the Model Context Protocol and establishing the Agentic AI Foundation ↗

Establishes the donation date; adoption figures are attributed to Anthropic.

[5] Foundation announcement · The Linux Foundation · Dec 9, 2025

Linux Foundation Announces the Formation of the Agentic AI Foundation (AAIF), Anchored by New Project Contributions Including Model Context Protocol (MCP), goose and AGENTS.md ↗

Confirms the receiving foundation and its initial projects; an institutional record, not an adoption audit.

Brief history

2026-09-05 · Initial brief separates the 2024 introduction from the 2025 foundation donation and explains the protocol’s boundaries.

2026-09-05 · Cross-checked the governance transition against the Linux Foundation’s own announcement and clarified what the institutional record establishes.

Suggest a correction.

Keep exploring