Technology publisher Contexte Tech has opened a Model Context Protocol server intended to let compatible AI assistants search its catalogue and retrieve source-attributed material directly. The launch is also a practical demonstration of a broader proposal: publishers and public-data providers can expose their own search and reading tools rather than waiting for a centralized search engine to copy and rank their content.

MCP is an open protocol through which an assistant can invoke capabilities supplied by an external server. Those capabilities can include searching a collection, reading a resource or completing a structured template. Contexte Tech argues that a single implementation can serve multiple MCP-compatible assistants and developer environments, reducing the need to build a separate integration for each client.

The publisher’s server exposes a catalogue of datasets, prompts, contexts and tools, much of it assembled from European public sources. One example described in the launch material uses the official French INSEE classification, published through data.gouv.fr, to help identify a business activity’s NAF or APE code. The assistant can load that dataset through the connector and identify the underlying source when presenting a result.

That source relationship is central to the project’s pitch. Conventional web search generally builds an index from copies of pages and then ranks those copies. In the model proposed by Contexte Tech, an assistant asks the originating server at query time. A statistics office, laboratory, public administration or specialist publication could therefore retain its own data, licensing terms and identity while delivering updated answers.

The approach does not guarantee reliable AI search by itself. Every participating server still needs clear provenance, stable access controls and honest descriptions of what its tools return. An assistant also needs to preserve the attribution supplied by the server and distinguish retrieved material from its own inference. Contexte Tech presents those requirements as unfinished implementation work rather than problems already solved by the protocol.

The article accompanying the server traces MCP’s introduction to Anthropic in November 2024 and says stewardship moved in December 2025 to the Agentic AI Foundation under the Linux Foundation. It also points to the protocol’s dated revisions and public registry as infrastructure for interoperability. Those historical and governance details are supplied by Contexte Tech’s account and were not independently corroborated in the packet.

For publishers, the release offers a concrete alternative to treating AI discovery solely as a licensing dispute with model and search companies. A direct server can define the available operations and attach source information to its responses. Adoption will determine whether that becomes a meaningful discovery channel. For now, Contexte Tech’s connector is a working example of decentralized retrieval, not evidence that MCP has replaced the web or conventional search.