Review and UGC apps with MCP servers: Idukki vs Okendo vs Tolstoy vs EmbedSocial

MCP is the Model Context Protocol. It lets an AI assistant call tools in another app. For a review or UGC app, that means Claude or ChatGPT can query your reviews directly. You skip the CSV export.

The four servers here answer different questions. Okendo reads public reviews. EmbedSocial reads your feeds and widget clicks. Tolstoy runs an agent that makes and publishes media. Idukki reads posts, products and sales, and can tag products if you allow it.

How we compared them

• Sign-in. No auth, an API key, or OAuth through the client.
• Read or write. Can the assistant change anything in your store?
• Clients. Which AI apps the vendor documents.
• Plan. The plan the vendor says you need.
• Data. What the assistant can see.

Facts come from each vendor's own help page or blog, read 13 October 2026. We make Idukki.

The four MCP servers side by side

Okendo MCP: public reviews, three tools

Okendo's server has three tools. The reviews_aggregate tool returns totals, rating spread, averages and recommendation share. The reviews_search tool filters reviews by text, rating, date and media. The questions_list tool returns approved Q&A for a product.

You switch it on under Settings, then Integrations, and copy a store URL. In ChatGPT and Claude you add it as a custom connector with no authentication. That works because the data is public. Okendo says the server only shows published, storefront-visible reviews and Q&A.

It needs an active Okendo Reviews Growth plan and admin access. It cannot read orders or order value.

Best for: an Okendo merchant analysing review themes in Claude. Limits: read only, reviews only, and anyone with the URL can query public data. See Idukki vs Okendo.

Tolstoy MCP: an agent that makes and publishes

Tolstoy's connection gives your assistant the Tolstoy agent and your media Library. It signs in with OAuth 2.1 and PKCE, so there is no key to paste. Its guide covers ChatGPT, Claude, Cursor and Codex CLI.

The agent can create and revise images and videos. It publishes media to your store, Meta Ads, Instagram, TikTok and X. It reads store, ad and AI widget analytics, builds AI widgets and tags products on videos.

Best for: a team that wants to make and ship video from a chat window. Limits: the guide names no read-only mode and no plan. Ask how publishing is approved before you connect. See Idukki vs Tolstoy.

EmbedSocial MCP: read your feeds and widget clicks

EmbedSocial's server is read access to your own account. Name a social account to pull its posts, UGC and reviews. Name a widget to read post, icon, lightbox and CTA clicks. Name a source to fetch media.

You add it in Claude or ChatGPT as a custom connector, then click Connect. EmbedSocial says it works through your API key and sees what you see.

Best for: a team reporting on review and social widgets. Limits: read only, and its MCP page names no plan. Its pricing page lists the API on Premium. See Idukki vs EmbedSocial.

Idukki MCP: read your sales, write only if you say so (ours)

We make Idukki. Our server has 12 read tools. They cover posts, products, albums, galleries, conversion, engagement, product and post analytics. One tool finds UGC for a product. Another identifies the products shown in a post.

There are three write tools. They tag a product on a post, untag it, or switch a gallery on or off. A read-only key does not see them. Every write attempt goes in the audit log with the key that made it.

You can connect two ways. Claude Code, Cursor and Claude Desktop take an API key starting idk_mcp_. Idukki stores only its hash. The claude.ai and ChatGPT connector screens sign you in with OAuth and PKCE. The consent page names the app, its return address and the business. Write access is a separate box, unticked by default.

A second, public MCP endpoint serves shopping agents. It has five read tools over rights-cleared UGC, reviews and ratings. The MCP page explains both.

Best for: asking an assistant which posts sold, then tagging products from the chat. Limits: three write tools only. The protocol version is 2024-11-05.

What to ask a vendor before you connect an AI assistant

1. 1What can it read? Public reviews only, or orders and analytics too? Ask for the tool list.
2. 2Can it write, and is that off by default? Publishing and tagging should need a separate grant.
3. 3Which workspace does it connect? With several stores, check the sign-in names the right one.
4. 4How do I revoke it? Find the screen that lists connected apps or keys.
5. 5Is there an audit log? You want to see which connection changed what, and when.
6. 6Who else can call it? A server with no auth answers anyone who has the URL.

For the shopper side of the same protocol, read MCP, tool calls and product feeds. For AI engines reading your reviews, see the best AI visibility tools for reviews and UGC.

Weighing MCP against feeds, rights and page weight? Our guide to choosing a UGC platform for AI shopping covers all eight questions.