Exclusive: LucidLink launches MCP server to give AI agents shared access to distributed files

LucidLink Corp., the maker of a cloud network-attached storage system based on object storage technology, today extended its distributed file system technology into agentic artificial intelligence with the public beta release of a Model Context Protocol server that lets AI agents access shared files across clouds, on-premises systems and edge environments.

The company said its LucidLink MCP server connects any MCP-compatible agent or orchestrator to a LucidLink filespace. The goal is to give multi-agent systems a persistent, writable layer with a shared state so agents, applications and humans can work from the same files without repeatedly copying or moving data.

LucidLink said the data movement problem is becoming more urgent as enterprises go beyond single-agent demonstrations into production workflows involving multiple agents and human reviewers. In those settings, the company said, the problem is no longer simply connecting an agent to a tool or data source, but preserving context, outputs and working state across sessions, nodes and frameworks.

“For the past 10 years, we’ve been solving distributed data challenges for teams who had to collaborate on shared assets,” said co-founder and Chief Executive Peter Thompson. But as the company saw customers beginning to connect agents to the same systems used by distributed human teams, they needed “shared, persistent context” in files that often live somewhere other than where the agents were running.

The MCP server is intended to expose LucidLink’s existing distributed streaming file system via the protocol that is becoming the de facto standard for inter-agent communication. The server works with Anthropic PBC’s Claude, OpenAI LLC’s Agents SDK, the open-source LangChain, LlamaIndex and CrewAI frameworks and any others that are MCP-compatible.

Thompson said the underlying access pattern for agents is not fundamentally different from the way people use files. The difference is that agentic workflows depend on files as memory, context and output. “For an agent working on a specific task, its output becomes a markdown file that needs to be the context for another agent,” he said.

That creates a persistence problem when workloads span multiple locations or infrastructure environments. Data may sit on-premises, in multiple clouds or at the edge. Moving it into a separate AI platform can create latency, governance and compliance issues, particularly for companies in regulated industries.

“The biggest problem for these larger enterprises is that data exists in multiple places,” Thompson said. “Going, finding it, moving, consolidating it and then giving it access in a non-shared way breaks the pipeline.”

Optimized for large files

LucidLink’s technology was originally developed for distributed teams working with large files in media production, engineering and other data-heavy environments. The platform includes block-level streaming, global file locking and zero-knowledge AES-256 encryption. The company says it has more than 6,000 customers and manages more than 95 petabytes of data.

Those capabilities are now being positioned as infrastructure for agentic AI. Global file locking prevents conflicts when multiple agents write to shared files. Zero-knowledge encryption means neither LucidLink nor any cloud provider holds customer encryption keys. The company also says the same namespace can work across the three major hyperscaler clouds, on-premises and in air-gapped environments.

LucidLink does not present the MCP server as a replacement for vector databases, data lakes or distributed table formats such as Apache Iceberg. Thompson said vector databases address retrieval, while his company is focused on the file-based write path that preserves outputs and makes them available to the next agent in a workflow.

“The file is the context,” he said. For example, one agent might create a transcript from a video file, while another uses both the transcript and the original video as context for a later step. If the files are in a shared LucidLink filespace, the next agent can immediately access them subject to permissions.

Thompson acknowledged that agentic workflows are still rare in enterprise environments. “A few customers are absolutely cutting-edge, but many others are just trying to figure it out right now,” he said.

Still, he said state management is already emerging as a practical obstacle. “If they don’t get that right, they’re not getting the output that they’re expecting,” he said. “They’re also finding that there’s just a lot more overhead in making it all work.”

Image: SiliconANGLE/ChatGPT