Tolstoy MCP lets an AI assistant connect to Tolstoy through approved tools. After you connect it, you can ask your assistant to work with Tolstoy from chat instead of switching back and forth between apps.

MCP stands for Model Context Protocol. It is a standard way for AI assistants to connect to external products through approved actions.

There are two Tolstoy MCP connections:

Use Studio MCP when you want to create or iterate on AI-generated assets. Use Library MCP when you want to find or organize existing media.

Think of the AI assistant as the operator and Tolstoy as the platform it can operate.

You give the goal in normal language. The assistant maps that goal to an available Tolstoy action, such as generating a Studio asset, refining the latest Studio draft, searching your library, inspecting an asset, renaming an asset, or marking an asset as favorite. Tolstoy performs the action, then the assistant summarizes the result and helps you decide the next move.

You do not need to memorize internal MCP tool names. Write the request the way you would brief a teammate:

For creative work, do not try to control every pixel in the first prompt. Give a clear brief and then refine the result.

For library work, be more exact. Include the asset name, creator handle, product, campaign, or specific item you want updated.

Open the Tolstoy platform and click the cable icon near the lower-left workspace controls. This opens the MCP setup area. Choose whether you want to connect Studio or Library.

Use Studio MCP when you want the assistant to generate and iterate on Studio images or videos. Use Library MCP when you want the assistant to browse, search, inspect, rename, or favorite media library assets.

Next, choose the setup guide for your AI client:

Try this first prompt:

Use Studio generation for a brand-new creative direction. The assistant sends your brief to Tolstoy Studio and returns the generated image or short video when it is ready.

Best for:

Use Studio iteration when you want to continue from the most recent Studio result in the same conversation. This keeps the context from the prior generation, so you can ask for changes without restating the entire brief.

Best for:

The media library actions help the assistant find and organize existing Tolstoy assets. It can list recent assets, search imported library assets by name or creator handle, inspect details, rename assets, and favorite assets.

Use listing when you want discovery. Use search when you know what you are looking for.

Widgets are Tolstoy experiences that can present shoppable, interactive, or campaign-specific video content.

Current MCP tools do not create, edit, publish, or check widget status. Use the Tolstoy platform directly for widget setup and publishing. You can still ask your assistant to help plan the widget, write copy, prepare creative prompts, or identify assets that might be useful for the widget.

Use planning mode, or the planning mode available in your AI assistant, when the work needs alignment before actions are taken. Planning is best for larger campaigns, unclear requirements, multi-step workflows, documentation updates, or anything where you want to approve the approach first.

Use normal execution when the action is clear, low-risk, and ready to run.

Strong prompts tell the assistant what success looks like. Include the brief, format, constraints, and output expectation.

Use this structure:

For creative prompts, include:

For library prompts, include:

For widget planning prompts, include:

Use planning prompts when the task is broad:

Start with one strong draft. Once the direction is close, ask for variants.

Give the assistant the job, audience, and channel. A prompt for a TikTok hook should be different from a prompt for a product page widget.

Include aspect ratio early. Use 9:16 for short-form social, 1:1 for square feed assets, 4:5 for portrait feed ads, and 16:9 for landscape web or YouTube placements.

Use constraints as plain rules. For example: "avoid medical claims," "do not change product color," "keep text under six words per frame," or "do not use celebrity likenesses."

Iterate instead of restarting. If a draft is close, ask the assistant to keep what works and change only what is wrong.

Use planning mode before spending generation credits or updating important library assets. Ask for deliverables, prompts, review steps, and acceptance criteria first.

Show matches before updating assets when names might be ambiguous.

Use the Tolstoy platform directly for widget creation, widget publishing, widget placement, and widget troubleshooting.

A request like "make a full launch campaign with all ads, update the widget, and organize the library" is too broad for normal execution. Start with a plan, then execute asset by asset.

"Make an ad" is less reliable than "make a 9:16 TikTok-style product video for mobile shoppers."

"Help with a widget" is less reliable than "plan a product page widget for the holiday bundle that uses one short video and drives shoppers to add the bundle to cart."

If the draft has the right scene but the wrong pacing, ask for a targeted edit. Restarting can lose the useful parts of the previous output.

If claims, text, brand rules, product accuracy, placement, or CTA matter, include them in the first prompt.

Library search works best with exact phrases or creator handles. If you are unsure of the name, list recent assets first.

When multiple matches are possible, ask the assistant to show results before renaming or favoriting anything.

Current MCP tools do not create, edit, publish, or check widgets. Use the Tolstoy platform for those actions.

Return to the Tolstoy platform's Settings -> MCP area and confirm you followed the setup guide for the correct AI client. Check that authentication completed successfully and that the connection is enabled.

Start a fresh login flow and make sure the browser opens the same account you use for Tolstoy. If your organization uses workspace controls, confirm that your account has permission to use the Tolstoy MCP.

Make the request explicit: "Use Tolstoy Studio to generate..." or "Use Tolstoy Library to find..." Also state what result you expect back.

Some clients show progress or generated assets differently depending on their MCP support. If the assistant cannot check status separately, open Tolstoy Studio directly to confirm whether the job is still processing or ready to review.

Turn vague feedback into direct instructions. Instead of "make it better," say "make the product larger in the final shot, reduce background objects, and slow down the camera movement."

Try a shorter phrase, remove punctuation, search by creator handle, or list recent assets.

Ask the assistant to show the matches with creation date, thumbnail, media type, or creator handle before taking action.

Library MCP is for finding and lightly managing existing media. Use Studio MCP for generation.

Studio MCP is for Studio generation and iteration. Use Library MCP for listing, searching, inspecting, renaming, or favoriting library assets.

The requested action may not be available in the current MCP connection. Check whether you connected Studio MCP, Library MCP, or both. If the request is for widgets or a standalone status check, use the Tolstoy platform directly.

Use planning mode to define the full campaign before generation or library updates start.

Create one approved direction, then generate variants around controlled changes: hook, setting, product angle, pacing, and CTA.

Use recent asset listing first, then update selected assets only after review.

Ask the assistant to summarize what changed after each iteration, then choose the next edit based on the review criteria.

Use the assistant to plan the widget experience and prepare creative assets, then use the Tolstoy platform for the actual widget setup.

Usually no. Say "Use Tolstoy Studio" or "Use Tolstoy Library" and describe the outcome. The assistant should translate that request into the relevant available Tolstoy action.

Use planning mode when the task is broad, expensive, ambiguous, or multi-step. Use normal execution for clear generation, search, rename, favorite, and simple library actions.

Yes. In the same conversation, refer to the previous result naturally: "make this warmer," "try 16:9," or "create three variants."

Not through the current MCP tools. Use the Tolstoy platform for widget creation, editing, publishing, and placement setup.

The assistant can list recent assets and search imported library assets by display name or creator handle. Search works best with close phrase matches.

Yes. Library MCP supports light asset updates, such as renaming an asset or toggling favorite status.

The assistant should ask before making changes when there are multiple possible assets, unclear intent, or a risk of updating the wrong item.

In the Tolstoy platform, go to Settings -> MCP.

Use the gallery section to show real examples once your team has approved them for documentation. Include the prompt, output, and the follow-up edit that made the result better.

Recommended gallery format:

Before publishing gallery items, verify that every example uses approved brand assets, avoids sensitive customer data, and reflects a workflow customers can reproduce.