How to Connect Tolstoy MCP to ChatGPT, Claude, Cursor, and Codex CLI

Tolstoy MCP is most useful when it stops being a concept and starts becoming the control layer for real ecommerce work.

If your team already uses ChatGPT, Claude, Cursor, or Codex CLI, the setup path is short. Tolstoy’s public MCP docs already expose the exact endpoints, OAuth behavior, and first workflows you can run once the connection is live.

This guide is the practical follow-up to our broader explainer on what MCP means for ecommerce. Instead of defining the protocol again, it shows how to connect Tolstoy MCP to the AI client you already use and what to test first once you are in.

Why this setup matters for ecommerce teams

The point of connecting Tolstoy MCP is not to add another technical integration for its own sake. It is to let your AI client work inside the same systems your team already uses to create and ship commerce content.

According to Tolstoy’s official MCP overview, a connected client can generate Studio content, manage your media library, build and publish shoppable widgets, tag products, and review ad performance right from chat. That changes the workflow from “AI gives me a suggestion” to “AI helps me move the work forward.”

For Shopify brands, that matters most when the bottleneck is operational: too many assets, too many surfaces, and too much copy-pasting between tools. Tolstoy MCP is useful because it turns that content layer into something an agent can act on.

Choose the right Tolstoy MCP server first

You do not always need both servers on day one. Start with the workflow you actually want to run this week.

Tolstoy Library

Use the Library server if you want to work with your existing commerce assets and storefront surfaces. The official Tolstoy Library docs list tools for:

• listing, searching, and updating media assets;
• creating and updating shoppable widgets;
• tagging products on videos;
• reviewing connected stores; and
• checking Meta ads performance or pushing assets into Meta Ads Manager.

If your first job is “find the right asset, make it shoppable, and publish it,” start with Library.

Tolstoy Studio

Use the Studio server if your first job is content generation. The Tolstoy Studio docs expose two core tools: generate_studio_content and iterate_studio_content.

That makes Studio the better first server when the work starts with “create a new image or video for this product launch,” not “organize and publish something that already exists.”

How to connect Tolstoy MCP in ChatGPT

ChatGPT is the fastest path because Tolstoy Library is already a published ChatGPT app. Tolstoy’s official Connect a client doc recommends this setup:

1. Search apps in ChatGPT for Tolstoy Library.
2. Click Connect and sign in with your Tolstoy account.
3. Ask ChatGPT about your library, widgets, products, or ads once the connector is live.

If you also want Studio in ChatGPT, Tolstoy documents a second path: enable Settings → Apps & Connectors → Advanced → Developer mode, then add a connector using the Studio endpoint.

The main reason to start in ChatGPT is speed. You can get to a live Library workflow without editing config files or adding a custom remote server manually.

How to connect Tolstoy MCP in Claude and Cursor

Claude and Cursor both support remote MCP connections, but the user experience differs slightly.

Claude

The official flow is straightforward:

1. Open Settings → Connectors in Claude.
2. Add a custom connector named Tolstoy Library or Tolstoy Studio.
3. Paste the correct endpoint and click Connect.

Once you sign in, Claude can call the same Library and Studio tools documented in the developer hub.

Cursor

Tolstoy’s docs keep the Cursor instructions intentionally short: add a remote MCP server in Cursor’s MCP settings with the correct endpoint, then let the first tool call open the OAuth sign-in tab.

That detail matters. Cursor does not need an API key here either. The first real MCP call triggers the Tolstoy login flow, and the connection is then tied to the Tolstoy workspace you authorized.

How to connect Tolstoy MCP in Codex CLI

Codex CLI is slightly different because Tolstoy’s docs note that codex mcp add targets STDIO servers, not remote HTTP servers. For Tolstoy MCP, you add the remote servers directly in ~/.codex/config.toml, then run the OAuth login command.

[mcp_servers.tolstoy-library]
url = "https://apilb.gotolstoy.com/mcp/v1/library/mcp"

[mcp_servers.tolstoy-studio]
url = "https://apilb.gotolstoy.com/mcp/v1/mcp"

Then log in:

codex mcp login tolstoy-library
codex mcp login tolstoy-studio

The first login opens a browser tab so you can sign in with your Tolstoy account. After that, the token is cached for future sessions.

If your team runs content or widget workflows inside Codex already, this is the cleanest setup because it keeps Tolstoy inside the same CLI environment where you are planning and executing work.

What to do right after you connect

The best first test is one workflow that proves the connection is useful, not just technically valid.

If you connected Library first

Follow the pattern Tolstoy documents in the Library guide:

1. Search for an existing asset.
2. Search for the relevant product.
3. Tag the product on the video.
4. Create or inspect a shoppable widget.
5. Publish the widget and follow the app-embed steps if needed.

This is the fastest proof that the client can actually help with commerce execution, not just content lookup.

If you connected Studio first

Start with one contained generation prompt. For example: generate a product image or video for a single launch asset, then refine it once with iterate_studio_content.

Tolstoy’s Studio docs also note that generated Studio content lands back in your Tolstoy library. That means the next step can be a Library workflow: tag the asset, place it in a widget, or push it toward your ad pipeline.

If you want that full loop in one platform, the public AI Studio and AI Player pages show where those outputs can go next.

Common setup mistakes to avoid

• Connecting the wrong server firstIf the task is asset management or publishing, start with Library. If the task is creation, start with Studio.
• Expecting an API key flowTolstoy uses OAuth 2.1 + PKCE. If you are looking for a static key, you are following the wrong setup model.
• Forgetting workspace scopeThe connection is bound to the Tolstoy workspace you authorize. Make sure you are signing into the correct account before testing.
• Stopping after connectionA successful OAuth login is not the real milestone. Run one asset, widget, or generation workflow immediately so the connector becomes part of work, not shelfware.
• Confusing “widget live” with storefront visibilityThe Library docs explicitly note that a widget set live in Tolstoy may still require the Shopify app embed to be enabled in the theme. Publishing inside Tolstoy is not always the final storefront step.

When to use Tolstoy MCP instead of a generic AI workflow

Use Tolstoy MCP when the job needs access to your real Tolstoy workspace: assets, widgets, product tagging, content generation sessions, or ad-performance context.

Use a generic AI workflow when the job is still open-ended strategy, naming, messaging, or brainstorming. The handoff point is simple: once the task needs to touch the actual commerce system, connect Tolstoy MCP and move the work into tools.

That is also why this setup guide belongs alongside the broader Tolstoy MCP page. The page explains the platform story. The connection guide makes the platform usable inside the AI client your team already prefers.

If you want to put this into practice, the best first move is to connect Tolstoy Library in the client you already use most, run one real workflow, and then add Studio once content generation becomes the next bottleneck. If you are not in Tolstoy yet, get started free.

FAQs

Do I need both Tolstoy MCP servers?

No. Start with Library if your first task is managing or publishing existing assets. Start with Studio if your first task is generating new content.

Does Tolstoy MCP use API keys?

No. Tolstoy’s public docs specify OAuth 2.1 + PKCE for every MCP connection, so you sign in with your Tolstoy account instead of pasting an API key.

Can I connect Tolstoy MCP to ChatGPT?

Yes. Tolstoy Library is already published as a ChatGPT app, and Tolstoy’s docs describe a direct connect flow inside ChatGPT.

What can I do with Tolstoy Library after connecting?

You can search and manage assets, inspect or create shoppable widgets, tag products on videos, review connected stores, and check Meta ads performance.

What can I do with Tolstoy Studio after connecting?

You can generate new images or videos and iterate on those generations inside the same Studio session before moving the output into the Tolstoy library.

What is the first workflow worth testing?

A good first test is one end-to-end Library workflow: find an asset, tag a product on it, inspect or create a widget, and publish it. That proves the client is useful for real commerce execution.