Let Your AI Agents Work on Your Files, With a Key You Control
Almost every team now has an AI assistant open in another window: Kiro, Claude, Cursor, or a script someone wrote on a Friday afternoon. And almost every team has the same small ritual around it. Find the right document. Download it. Paste it into the chat. Wait. Copy the answer back. Upload the result. It works, but the assistant never sees your actual project, only the scraps you hand it, and every scrap leaves a copy somewhere you did not intend. Today we are removing that ritual: LightUp.Cloud can now give each of your AI agents its own key to the files you choose, and nothing more. The agent connects through the LightUp Cloud MCP, a small connector for the open Model Context Protocol that Kiro, Claude Desktop, Cursor and many other agents already understand.
A Monday morning, before and after
Before. A landscape designer asks an assistant to draft the planting schedule for a courtyard. She exports three PDFs, hunts for the surveyor's latest notes in a folder she half remembers, pastes it all into a chat window, and then discovers the notes were two versions old. The schedule is fine; the afternoon is gone.
After. Her assistant has the LightUp Cloud MCP configured, so she types the same request and nothing else. The assistant searches the courtyard project in LightUp, in the spaces her key allows, for the surveyor's notes, the client's brief and the plant list. It checks the version history to be sure the notes are the latest, and writes the draft schedule as a file into the sync folder on her own computer. Her LightUp sync app, which she already runs, uploads that file to the project space it is linked to, so the draft appears there like any other file: she opens and corrects it, her colleagues see it, and any agent that has access to that space can find it later. No exporting, no pasting, no stale copies.
What your agents can do with LightUp Cloud MCP
Add the LightUp Cloud MCP to any agent you use, and that agent can work with your files in the spaces you allowed it:
- Find files by what is in them, not only by their names: titles, descriptions, keywords, transcripts of recordings and the text and content that LightUp's indexing has found in documents and images. Phrases like "site plan with retaining wall" work. The search runs inside your LightUp storage; the agent receives only the results.
- Read a file, and see its version history and the recent changes in a space: who changed what, and when.
- Lock a file before editing it, so a colleague or another agent cannot overwrite the same file at the same moment, and unlock it when done.
- Write new or changed files. The agent saves them into a folder on your computer that LightUp synchronises; your LightUp sync app then uploads them to the space that folder is linked to, where you, your colleagues and other agents with access to that space can open them, with version history as for any other file.
Deleting files and restoring old versions exist too, but they are off unless you grant the permission and switch those tools on. An agent that was never meant to delete simply has no way to.
Who gains what, with examples
Anyone whose work is a pile of files and a deadline. The scenarios below are illustrations of how the pieces fit together, not customer stories, and the time you save will depend on your team. One point applies to all of them: big files stay where they are. An agent reads text and small files directly (up to a size limit you can set), but a RAW photo, a camera original or a large drawing is found through what LightUp has indexed about it (descriptions, keywords, transcripts, text inside) and is left untouched in storage. What the agent writes beside them is small and useful: lists, logs, summaries and drafts.
Developers
Keep the knowledge of the project where every agent can reach it. A team stores its documents in LightUp: specifications, design notes, user guide drafts. They synchronise to the server the team controls, which can be a machine in their own office. From then on any agent a developer prefers, once the LightUp Cloud MCP is configured, shares that common knowledge. There is no tool to build, and no need to hand sensitive documents to a third-party cloud service: the documents stay in your storage, the search runs there, and only the few passages the agent fetches for a task reach the model you chose.
An example: a team is building a diagnostic tool for a telecommunications operator, and the draft user guide describes the test procedures, the error codes and the report formats. A developer asks the agent to implement the module that reads line-test results. The agent searches the guide through the MCP, fetches the sections on the test procedures and the error codes, and uses them while writing the code, naming the files and sections it relied on. When the guide is revised, the next question is answered from the new text.
Architects
An architect's models live in CAD or Lumion on the workstation, and what leaves the office is a rendered image in a PDF, often sent to the customer over WhatsApp. The customer's own PDFs, used as input, sometimes carry hidden metadata (names, software, file paths), and a bureau that forgets to remove it can find itself with a confidentiality problem. LightUp can show the embedded metadata of a PDF and erase it in one click (we wrote about how that feature was born). With an agent in the loop, the check becomes part of the routine instead of something remembered at six in the evening: before a delivery the architect asks the agent for every PDF in the project's outgoing folder and the customer files it was based on, and gets one list of what is about to leave the office, so each file can be cleaned in LightUp before it is shared.
Landscape designers
After the courtyard is built, the client's gardener needs to know what to do and when. The designer asks the agent for a maintenance calendar. It finds the final planting schedule and the site photographs of the north slope taken across the seasons, and writes a month-by-month care calendar as a draft beside them, which the designer reviews before it goes to the client.
Photographers
After a wedding there are four thousand frames and a couple waiting. The photographer describes what she needs in words: "the first dance", "grandmother laughing near the window". The agent searches the indexed descriptions and returns the matching file names, then writes a shortlist file per request. She opens the shortlist, not the whole folder. The RAW originals are never copied into a chat window.
Video: editors, cinematographers and script writers
A film project is many kinds of file at once: scripts, shot lists, camera reports, lighting notes, interviews and footage. Pre-production, including building the world, casting, writing the shot-by-shot script, storyboards and the approval of every material, happens in LightUp Studio, our AI film-production suite. Around Studio and the shoot, the agent helps each role with the files it keeps in LightUp.
- Video editors have six hours of interviews. The editor asks where the contributor first mentions the flood; the agent searches the transcripts, returns the clips and the passages, and writes a logging sheet with file names and quoted lines in the edit folder. A lock keeps anyone from overwriting the cut notes in the meantime.
- Cinematographers ask which scenes of the approved shot list still lack a close-up, and which camera reports have no lens recorded. The agent reads the paperwork and the indexed descriptions of the footage (it does not pretend to watch the originals) and saves a coverage report beside the shot list.
- Script writers ask the agent to read the latest draft and list continuity problems: a character who has a coat in scene 12 and not in scene 13, a date that does not add up. The notes are saved in a separate file next to the script, never over it, and the version history shows who changed which scene.
Small and medium businesses with big files
Companies of ten or fifty people rarely have an IT department to build integrations, and they often work with the heaviest files: drawings, scans, footage, print masters. A connector that works with the agents they already use, on files they already have, suits them best.
- A marketing agency keeps campaign assets, brand guidelines and client briefs in LightUp. An account manager asks for everything approved for the spring campaign and gets a hand-over list with file names and dates, ready to send to the client, in minutes rather than a morning of opening folders.
- A survey and engineering firm stores drawings, reports and point-cloud exports for dozens of sites. A project manager asks for all deliverables for one site and gets a checklist of what exists, what is missing and which version is current, which makes invoicing a milestone a matter of checking a list.
- A small manufacturer keeps CAD files, specification sheets and quality reports. When a supplier asks which revision of a part is current, the agent finds the specification, reads the version history and answers with the revision and who approved the change, while the CAD model stays in storage.
What it means for you
In plain terms: you spend less time looking for files and moving them around. Your assistant answers from the files as they are today, not from whatever you remembered to paste. What it prepares lands in the shared space, where your colleagues see it and the version history keeps every change. Your documents stay in the storage you control instead of being copied into chat services. And you decide, agent by agent, which spaces it may use and whether it may only read or also write.
Your public showcase, found by the new readers of the web
More and more people ask an AI assistant instead of typing a query into a search box. If you publish a showcase (a portfolio, a project page, a book), those assistants are now among its readers, and we do our best to help authors be found by them. A LightUp showcase is public by design, and we publish it together with the machine-readable information that AI agents on the internet look for: a public JSON feed of your showcases that needs no login, announced in our llms.txt file. This lets assistants learn about your projects, quote them correctly, and credit you with a link back.
You stay in control. Per page, you can switch off the AI-friendly feed, and you can state whether your content may be used for training models; that is off by default. Only staff and administrators of the team can change these two switches.
With the LightUp Cloud MCP, your files become a workspace that AI agents can use on their own, so that they can look up, prepare and hand over the routine parts of your work and orchestrate it between themselves, while you decide what they may reach.