Part 1: connecting the Sitecore Marketer MCP server to VS Code, and using it to build and check a multilingual page from plain prompts.
The Sitecore Marketer MCP server is a remote MCP server that lets an AI client work with your SitecoreAI tenant through natural language. Behind the scenes it calls the Agent API, so a prompt like "create a landing page" turns into real items, components and datasources in your site.
A few months back I built a small no-code prototype on SitecoreAI for an RFP. Five pages, a handful of reusable components, English and Spanish. The pages came together quickly. The Spanish version took far longer, mostly because of one trap: translating the page is not enough, every component datasource needs its own language version too. So when I started playing with Marketer MCP, that was the first thing I wanted to test. Can it build the page and also tell me what is missing in the second language?
In this two part series, Part 1 covers Marketer MCP from VS Code. Part 2 covers building a workflow agent in Agentic Studio for the same job, so a content author can run it without an IDE.
How I tested this, On a new non-production SitecoreAI environment with one site (ai-playground) and two languages, en and es-ES. The server was registered in VS Code, and I ran the prompts from the Claude Code extension inside VS Code. Every tool name below comes from the permission prompts I saw during the run. The screenshots are from the same run.
Prerequisites
1.A SitecoreAI account with at least the User app role, and access to the organization and tenant you want to work with. What the server can do follows your Sitecore role.
2.An MCP client. I used Visual Studio Code. Claude Desktop, Cursor, Copilot Studio, Postman and the Codex app are also documented.
Step 1: Connect VS Code to Marketer MCP
Open the user profile mcp.json in VS Code (Command Palette, then MCP: Open User Configuration) and add the server:

Click Start above the server name. VS Code asks if the server may authenticate.

Click Allow, then Open. The browser shows the Marketer MCP authorization request.

Click Allow Access, sign in with your Sitecore account, and pick the organization and tenant. Every tool call after that is scoped to that tenant.credentials and then provide the tenant name


Back in VS Code the server shows Running with 62 tools.

Open the tools list in chat and you should see the Marketer tools, things like add_component_on_page, add_language_to_page and create_page.

One thing to know if you use the Claude Code extension as the chat client, as I did. It does not read VS Code's mcp.json. It needs a .mcp.json file in the project root, and the top key there is mcpServers, not servers. I copied the file across with the wrong key first, and the server simply did not appear until I changed it.

Step 2: Find your way around the tenant
Start with something read-only. My first prompt was:
Prompt 1 :List the sites in this tenant, then show me all pages under the demo site and the languages available.
Tools called: list_sites, get_all_languages, search_site.

My first run returned no sites at all and only English. The environment was new and I had not created a site yet. After I created the ai-playground site and added Spanish in the site settings, the same prompt returned this:

Prompt 2 :Show all pages under the ai-playground site, and list the components I can add to the main placeholder of the Home page.
Tools :

Step 3: Create a page with components
Prompt 3 :On the ai-playground site, create a page called "Digital Banking" under Home. Add a Hero component if one is available, otherwise a Promo, with heading "Bank from anywhere", a short body text, and a link to /contact. Then add two more Promo components below it as cards, titled "Open an account" and "Talk to an advisor", each with one line of text.
Tools called: create_page, add_component_on_page, update_fields_on_item.

The starter site has no Hero component, so it used a Promo for all three and named the datasources Hero_Promo, Open_an_account and Talk_to_an_advisor under the page's own Data folder. Open the page in Page builder and check it. Nothing is published yet, which is how it should be.

Step 4: Add Spanish, and check what's really translated
Prompt 4 :Add a Spanish (es-ES) version of the Digital Banking page. Also add a Spanish version of the first component's datasource with the heading "Banca desde cualquier lugar". Do not translate the two cards.
This gives us a mixed state on purpose: one datasource translated, two not.
Tools called: add_language_to_page, get_content_item_by_id, update_fields_on_item.

Now look at the bottom of that screenshot. The Spanish version of the page is empty. No Promo, no cards, only an empty main placeholder. The three components had been added to the English version of the page, and the new Spanish version did not get them. The Spanish heading existed on the datasource, but there was nothing on the page to show it.
So this is a second trap on top of the one I started with. A page can have a Spanish version and translated datasources, and still show nothing, because the components are not on that language version. I had to ask for it:
Prompt 4a:Add the same three components to the es-ES version of the Digital Banking page, in the same order and placeholder, reusing the existing datasources.

Prompt 5 :List every component on the Digital Banking page with its datasource, and tell me which datasources have no Spanish (es-ES) content.

It listed the three components with their datasource IDs, reported Hero_Promo as having es-ES content, and reported both cards as having no Spanish version. It also noted that the body text and the link on Hero_Promo were still English, which was correct.
I didn't want to take the model's word for it, so I opened each datasource in the Content Editor, switched to es-ES, and compared.
Datasource What MCP reported What I found
Hero_Promo es-ES version with the Spanish heading Matches
Open_an_account No Spanish version Matches
Talk_to_an_advisor No Spanish version. Matches

Prompt 6:Create Spanish versions of the two card datasources with titles "Abrir una cuenta" and "Hablar con un asesor" and a one line Spanish text each.
which, per the tool reference, uses update_content on each datasource in es-ES.

Step 5: Verify on Experience Edge after publishing
Marketer MCP writes to the CMS. It does not publish for you, and you shouldn't want it to. After publishing the page, I still check Edge directly with one query in the GraphQL IDE, because it lists every field and value on an item without me guessing field names:
Query :
query { item(path:"/sitecore/content///Home/Digital Banking/Data/Hero_Promo", language: "es-ES") {
name
fields {
name
value
}
}
}

My first attempt returned "item": null. I had typed the path without the leading slash. So read the result like this:
- item is null: the path is wrong, or the item is not published in that language. Check the leading / first.
- The item comes back but values are empty: the version is published, the fields are not filled.
- A specific field is missing: check the field name. They are case-sensitive.
What went wrong, in short
- The first prompt returned nothing because a new environment has no site. The model's explanation for it was wrong.
- There was no Hero component in the starter site, so I got a Promo. Ask what is available before you ask for a layout.
- The Spanish page version had no components until I asked for them separately. The translation check in Prompt 5 only makes sense after that.
- The Edge query returned null because of a missing slash in the path.
Things to keep in mind
- The server only does what your Sitecore role allows, and every operation is audit logged.
- Rate limit is 50 requests per 10 seconds. A long multi-step prompt can hit that, so break big jobs into smaller prompts.
- delete_content and remove_component_on_page check references and ask for confirmation first. Read those confirmations carefully.
- search_site matches page names only, not body text.
In Part 2, I will take the same page , check and build it as a workflow agent in Agentic Studio, so authors can run it from SitecoreAI itself.
Read More :
Top comments (0)