Komprise has launched Universal File MCP, a single interface through which AI agents and large language models can query enterprise file storage, NAS, cloud and unstructured data silos, replacing the stack of per-vendor MCP connectors that the company says degrades agent accuracy and inflates compute costs.

MCP, or Model Context Protocol, provides secure, bi-directional API-level communication between LLMs and agents and external data sources or tools. MCP servers handle requests from AI clients, process them, and return contextual responses. Nearly every storage and data vendor now publishes its own MCP interface, and Komprise argues the resulting multiplication of tool definitions overloads AI systems with lower accuracy, slower performance and higher costs.

The problem compounds with what Komprise calls unstructured data response bloat. File and object listings are unbounded and can span millions to billions of files across disparate hybrid storage. A McKinsey report cited by the company puts 60 percent of agentic AI computing costs down entirely to "response refinement" — the behind-the-scenes token loops where sub-agents critique and format data before presenting a final answer.

The practical effect: an organization sends one MCP request to Komprise rather than multiple MCP requests to individual data sources, and Komprise's software right-sizes responses by returning just the relevant unstructured data, enriched with context and governed by user-specific access permissions.

CEO Kumar Goswami positioned the product as an answer to dark data spread across supplier boundaries. "The Komprise Universal File MCP lifts the veil on dark enterprise data by unifying governed access across silos," he said. "We're thrilled to help our customers easily leverage AI across all of their file data by asking complex questions and initiating workflows to act on the results, all while ensuring access controls remain intact."

The architectural choice

The launch lands on a long-standing fork in unstructured data management. Applications that need to act on an organization's data estate typically face multiple separate data sources, and inspecting and correlating a data index across distributed silos takes time. Enterprises can consolidate everything into one supplier's silos — go all-in on NetApp or Dell, say — or use a third-party discovery and classification system that builds a single virtual index without moving data from the source silos. Komprise sits in the latter camp, providing cross-supplier, cross-silo access to distributed, multi-supplier data estates.

What the interface does

Universal File MCP ships with seven capabilities the company outlines:

• Governed, secure access. Authenticated access delivers responses based on the user's privileges, and the system audits what data went to AI for governance and reporting.
• Consistent schema. The petabyte-scale Komprise Global Metadatabase gives unstructured data a uniform structure regardless of which vendor produced the data or where it currently lives.
• High-quality, enriched data. Komprise AI Preparation & Process Automation (KAPPA) extracts contextual metadata based on industry, enterprise, sensitivity and users to support specific AI use cases.
• Noise filters. Komprise Deep Analytics lets users eradicate irrelevant, outdated, conflicting and duplicate data before it reaches the model.
• Progressive disclosure. The interface loads information to AI progressively, starting with metadata for summarization, with the option to export results as an Apache Iceberg table via Komprise Transparent File Tables. It loads files only when the AI needs them, further reducing response bloat.
• Tiered data access. Data stays reachable even as it moves, so tiered and archived content remains accessible through Komprise Transparent Move Technology.
• Ingest into AI, lakehouses and analytics. Follow-up actions from a prompt can push results into AI systems or a lakehouse using Komprise Intelligent AI Ingest.

Customer validation

One early customer frames the value in security terms. Stephen Clark, Director of Information Security at Children's Medical Center, Dallas, said: "Opening up clinical and research data to AI tools means my team needs to review who could see what, case by case, putting security in the middle of every request. With Komprise Universal File MCP, the query respects the permissions a user already has and leaves us a record of what went to the model. Our researchers get their data faster, and IT is not managing a separate connector for every storage system."

For Clark, the product solves the multiple separate connector problem outright.

Komprise Universal File MCP is available today through an early access program for customers and partners. As MCP servers proliferate across the storage industry, the launch signals that the competitive battleground in AI-era data management is shifting from connector count to how efficiently and safely each request returns the right data — and vendors that can shrink response bloat while preserving permissions stand to cut the token costs that now dominate agentic AI budgets.