Is MCP becoming the common foundation of smart, decentralised search?

The MODEL CONTEXT PROTOCOL (MCP) is an open standard that lets AI assistants call publisher-hosted tools - search, resource readers and templates - so answers come directly from the original sources instead of a frozen model memory or a single central index. Launched by Anthropic in November 2024 and ceded to the Agentic AI Foundation under the Linux Foundation in December 2025, MCP is versioned and publicly registered and already works with major assistants and developer tools. Under MCP, each publisher runs its own server, assistants query multiple servers on demand, and responses carry metadata (author, licence, source, last update). That decentralised flow promises up-to-date, verifiable, licence-aware answers and reduces the incentive for models to hallucinate by making missing data explicit.

Practical gaps remain: servers must enforce machine-readable citations, treat content strictly as data (not as executable instructions), and provide privacy-preserving measurement so publishers know how often assistants refer to them. Contexte Tech’s MCP server demonstrates these specifics: an open catalogue (no account or key) offering multilingual search, resource loading by short code, server-required citation lines, assistant-identifying but non-personal tracking links, and flags for protected content. A quick test is adding the server as a connector and asking it to load n8qhje, which returns an INSEE-based NAF code dataset with full source citation. Widespread, honest MCP servers aim to become the common foundation for smart, decentralised search and a practical path for publishers to surface authoritative data to assistants.