# LucidLink giving AI Agents file access with MCP > Source: > Published: 2026-06-29 13:10:16+00:00 # LucidLink giving AI Agents file access with MCP [LucidLink](https://www.blocksandfiles.com/object/2026/03/10/lucidlink-connect-streams-s3-buckets-without-the-ingest-headache/5208840) is giving AI agents the same access to real-time, streamed file data as its users and applications, and storing state across and between agents in a workflow. LucidLInk’s Filespaces product provides a cloud-based object store with a filesystem layered on top. It mounts as a local storage as far as remote users and applications are concerned and they access files on it as if they are stored locally. But they are not. Instead the bytes are streamed in real time, thus avoiding file copying operations from distant sources. Data is encrypted which helps with security and sovereignty. Now LucidLink is thinking about AI agents accessing file data and seeing that its Filespaces product can serve them as well. CEO and co-founder Peter Thompson has [blogged](https://www.lucidlink.com/blog/agentic-ai-shared-data-layer) about this. He says agents forget their context between sessions and can’t hand it on to other agents in a workflow involving several agents. Adding CP server access to Filespaces enables both problems to have fixes. He writes: “LucidLink was built to access … remote data in place. Today, we’re opening that infrastructure to every agent pipeline, starting with the public beta of our MCP server.” He points out that “MCP is the protocol layer, not the persistence layer. It tells agents how to reach resources. It doesn't provide the durable, shared, writable state that persists across sessions, survives handoffs and stays consistent when multiple agents write concurrently from different nodes.” With the LucidLink MCP Server custoomers can “Connect any MCP-compatible agent or orchestrator to a LucidLink filespace and you get the persistence layer that multi-agent pipelines have been missing. …. an agent's working state is a folder in the filespace. The next session, a different compute node, a different framework version — the state is there. …one agent's output folder is the next agent's input. Handoffs are instantaneous. State lives in the filespace, where every participant sees exactly what they're authorized to see. … Claude Code, LangChain and CrewAI are among the frameworks that already work with LucidLink as the shared data layer underneath, giving agents persistent access to the same filespace teams already use. ” Human-in-the-loop working is supported: “Because LucidLink mounts as a local drive, a human can connect to the same filespace an agent is working from, supplying data, monitoring, dropping instructions and reviewing outputs, directly inside the agentic workflow.” There is auditing support: “Every write is traceable, every version recoverable, zero-knowledge encryption and global file locking on by default.” Thompson says the Filespaces “namespace works identically whether agents run on AWS, GCP, Azure or on-premises hardware. The agent doesn't know where the infrastructure lives. The data is just there.” The LucidLInk MCP Server public beta is open and interested parties can get started [here](https://www.lucidlink.com/mcp-server).