Anthropic's open Model Context Protocol (MCP) standard is letting vendors pipe continuously curated market and competitive data into Microsoft Copilot, Anthropic's Claude, and the agent tools enterprises are adopting. Helsinki based Valona is the latest to ship on it.
Nearly nine in ten organizations now use AI, according to Valona-cited McKinsey 2026 research. Most of those organizations are running the same foundation models against the same productivity tasks. The thing that cannot be commoditized, and the thing starting to separate genuine competitive intelligence from expensive autocomplete, is what those models have to work with.
That framing is the spine of a new industry thesis pushed this week by Helsinki-based intelligence vendor Valona, and echoed by a growing crowd of competitors, infrastructure providers, and protocol designers who all want to own the data layer that enterprise AI agents actually read from. The competitive advantage does not come from having AI, Valona's Chief Product Officer Stuart Reynish said in the company's June 16 launch announcement. It comes from what the AI has to work with.
The intelligence layer, in plain terms, is the continuously curated, validated, and refreshable feed of market and competitive data that an agent can call into a Copilot session, a Claude thread, or a custom workflow without re-prompting from scratch. Valona co-founder and CEO Eetu Laaksonen has called this the hardest part of enterprise AI to get right, and the part that decides whether a "live" intelligence system is a real capability or a marketing slide. Regenerating the same competitive analysis every time a question comes up is expensive, slow, and inconsistent across teams, and that cost is the one the new layer is trying to retire.
What changed this spring is the plumbing. The Model Context Protocol (MCP), an open standard originally introduced by Anthropic, has matured into a near-default way for AI systems to reach into external data sources. The acronym shows up more often in product roadmaps than in trade press, but the mechanism is straightforward: MCP lets a model request structured context from an external server the way a browser requests a page, and lets a vendor expose that context without rebuilding a custom integration per model. That portability is why vendors are lining up.
Valona, a twenty-year-old competitive and market intelligence provider that has been selling curated research to strategy teams since before the LLM era, is one of the latest to productize that layer. Its new MCP server pipes its proprietary market and quantitative financial data into Microsoft Copilot, Anthropic's Claude, and other enterprise agent frameworks, with the explicit pitch that customers should stop regenerating the same analysis from scratch on every prompt. The launch is one of several MCP server announcements in the enterprise intelligence category in 2026, and the framing is consistent across them: AI is now a commodity, the open question is what data and curation discipline sits underneath it.
The critique is fair, and it is worth holding in the open. "Validated, current, and ready to act on" is a vendor claim until an independent benchmark or a named customer proves it. "Always-on intelligence" is only as good as the last refresh cycle, and most enterprise AI "intelligence layer" pitches to date have been wrappers around a single analyst's spreadsheet with a chat interface bolted on. Valona's release identifies a select group of enterprise customers piloting the MCP server in Copilot environments but does not name them, discloses no pricing, and leans on a secondhand McKinsey figure that the primary report would need to confirm. None of that is fatal to the thesis. It is a reminder that the new layer is being sold before it is being audited.
For an enterprise reader trying to spend the next dollar of evaluation energy wisely, the practical framework is the same regardless of vendor: ask about data lineage and source provenance, ask how often the underlying feed is refreshed, ask what the latency is between a market event and a model's first chance to see it, and ask whether the vendor exposes its data through MCP or another portable protocol, or only through a custom integration that locks the customer into one model. Vendors that can answer all four crisply are doing something real. Vendors that cannot are selling a chat box.
The thing to watch next is whether the MCP standard itself stays vendor-neutral. Anthropic shipped it as open source, and the protocol is now being adopted by competitors who have no particular reason to be loyal to its original author's roadmap. If MCP holds together as a shared interface, the intelligence layer is genuinely portable and the model layer stays interchangeable, which is the world Valona and its peers are betting on. If the protocol fragments or gets quietly absorbed into a single hyperscaler's stack, the moat moves again, this time to whoever controls the connector.