MCP data and privacy

By connecting an assistant, you allow the AI client you use to read and change your Slicer projects within the access you chose. This page explains which data goes where.

What the agent receives#

Tool results end up in the conversation with the model, that is, with the company whose AI client you use. Depending on the tools called, this includes:

• balance, prices and account settings (no email or payment details);
• project titles, statuses, montage settings, wishes for the director;
• the video text — a word-by-word transcript, captions, hook, plaques;
• upload details: file names, duration, frame size;
• connected social accounts: platform and account name, publication statuses;
• the material for your own director (get_director_brief): the sources, speech by words, pauses, retakes and your wishes for the director — without the rules and lessons of the Slicer AI director: they stay on the Slicer server;
• frames of the sources — only when the agent asks for them itself (get_frames, Your own director): up to 16 images of 512 pixels per call, right in the tool result.

The videos, photos and audio as a whole are not passed to the agent, and it gets no links to files: the server returns only the frames the agent asked for and a link to the project in the app. Captions from speech are recognised by Slicer — just like in the app — and the agent receives only the text. How the AI client stores conversations and whether text is used for training is governed by its own policy — check your client's privacy settings.

What Slicer receives#

• Files the agent uploaded and projects it created are the same as ones created in the app: encrypted, available only to you, kept for the same periods. In the project list they're marked with their source (mcp).
• Slicer doesn't see your requests to the agent — only tool calls with their parameters.
• Slicer uses the agent's video plan and its observations from the frames (scenes) to build the video and doesn't store them separately: the project keeps the video and, from the plan, the title, description and hashtags for publishing.
• For each connection Slicer stores: the client, its return address, access, limit and last use time. Paid operations and edits are recorded in the project history, just like from the app.

Docs#

search_docs and read_doc search and read only public pages of docs.slicer.media — your data doesn't go into these requests.

How to limit access#

• Minimal access. For questions and browsing, the Read only preset is enough (with it the agent can also look at frames of the sources); to work without spending, remove montage:run; without publishing, remove publish.
• A separate connection for each client. Then disconnecting one doesn't affect the others.
• Disconnecting. Profile → Developers → Connected apps → Disconnect. Access stops immediately and the client's tokens stop working. Data the client already received in the conversation stays with it — delete the conversation in the client itself if needed.
• Deleting your account revokes all tokens and connections automatically.

More on encryption, retention and access by staff — on the Security page.

What's next#