Model Context Protocol (MCP)

The Model Context Protocol (MCP) is an open standard, introduced by Anthropic, for connecting AI language models and agents to external tools, data sources, and function calls through a uniform interface, so a model-calling agent can reach a given tool without bespoke integration code per pairing.

This vault first met the term on 2026-07-09, in a SANS Institute post framing MCP (alongside persistent memory and multi-agent handoffs) as one of the tool surfaces through which an AI agent's attacker-controllable natural-language input can be steered — the "confused deputy" problem revived for agentic systems (claim-confused-deputy-2026-ai-agent-security-frame). It resurfaces on 2026-10-27 in a very different register: a Babylonian-tablet dating pipeline uses MCP as the interface through which its agents query ephemeris lookups and eclipse-visibility catalogs against 701 candidate years (claim-babylonian-diaries-dating-pipeline-parser-llm-mcp-astronomy) — the same protocol as attack surface in one domain and as auditable plumbing in another.

References

• claim-confused-deputy-2026-ai-agent-security-frame
• claim-babylonian-diaries-dating-pipeline-parser-llm-mcp-astronomy