Announcing Sitecore Community MCP v2

Sitecore Community MCP 2.0: four API surfaces feeding one MCP server - SPE Remoting, Item Service, Edge GraphQL and the new Authoring GraphQL - with 121 tools, 22 of them for authoring, and response sizes cut from 777,501 to 3,950 characters for GraphQL introspection

I'm thrilled to announce a major release of the Sitecore Community MCP.

I use the Sitecore Community MCP every day. Some of my common uses:

Scaffolding new Sitecore components
Adding new fields to extend components
Content operations such as bulk updating pages or datasources
Content migration projects from other CMS's

I primarily use Claude to interact with the MCP, and I noticed that occasionally it would say things during those operations like "that tool call just blew through my context window", or I would notice it have to retry operations with different parameters. I realised that this meant there was a need for some improvements to be made on the MCP.

What's changed?

MCP V2 Spec

First of all, this MCP needed to be updated to the latest MCP v2 spec

The highlight of this release is a stateless protocol core - MCP is transforming from a bidirectional stateful protocol into a request/response stateless protocol. It was one of the most highly-requested features from developers who were eager to get better reliability and scalability for their MCP servers.

That was a relatively easy upgrade for this MCP as it already was stateless, but it helps refine the conventions. We are still backwards compatible to the older spec, the Readme has some nifty badges now which say what MCP specs are supported.

Support for the Authoring and Management API

Sitecore has shipped a powerful Authoring and Management GraphQL API in both XP and SitecoreAI for quite some time now. I believe it was in XP since the 10.1 days. This extra interface closes the gap a lot over what the Item Service REST API provides.

For those developers who have made their own SitecoreAI Marketplace apps, they may already be somewhat familiar with some of the capabilities of this API.

Covered some gaps with the Sitecore Marketer MCP

Sitecore's own Marketer MCP has some SitecoreAI specific marketer capabilities which are outside of the scope of our MCP, however it did have some page / content related tools that were useful, so we've added them in this release. The full list of added tools is too big to cover here so see the release notes if you're interested.

For a comparison:

	Sitecore Community MCP	Sitecore Marketer MCP
Platforms	SitecoreAI and XM/XP, all versions, including your local Docker CM	SitecoreAI only
Hosting	Self-hosted (npm, Docker, source); credentials never leave your infrastructure	Hosted by Sitecore; OAuth through the cloud
Surface	121 tools across the full developer surface: items, templates, presentation, media, security, indexing, logs, publishing, GraphQL, raw PowerShell	Marketer operations: pages, components, briefs, brand kits, personalization, A/B tests
Escape hatch	`run-powershell-script`, raw Edge GraphQL and raw Authoring GraphQL. If a tool doesn't exist, the capability still does	Closed tool set
Layout safety	`add-rendering-to-placeholder` refuses components the placeholder settings forbid, naming the allow-list	Writes are not validated against placeholder settings
Tool gating	`TOOL_GROUPS` / `DISABLED_TOOLS` / `TOOL_PROFILE` trim the schema cost per turn	Fixed tool list

I'll note that if you are playing in the marketer space (e.g. personalisation rules, brand kits) then you'll still want the Marketer MCP, no harm in having both available.

I was A/B testing the Community MCP against the Marketer MCP for building pages and found that the Community MCP gave better results. In fact my test found a bug in the Marketer MCP that it wasn't able to overcome:

Claude explaining that update_fields_on_item on the Marketer MCP returned an empty updatedFields and silently dropped values on Heading, Document Tile and Video components, and that the Community MCP authoring tools could finish them

Consolidated and optimised tools

Each tool that is available on an MCP adds a token cost on every turn that your AI agent takes. An effort has been made to optimise that by consolidating similar tools, for example instead of a separate get-item-by-path and get-item-by-id tool, those are consolidated to get-item and accept params for either option.

The tool descriptions and titles have been optimised too, tools that noone will ever use or need have been deleted, and further there is config for token-conscious devs to be able to selectively disable tools that they don't want in their context.

So even though we've added a whole lot of new tools, overall the v2 release uses about 30% less context by default.

Additionally, various tools were returning an unreasonable amount of data in their responses, some of it totally useless! Every byte that is returned by a tool counts towards your tokens so we want to save you cost and save your context window. I've added a layer of code to the server which strips out useless bytes from server responses and supports shaping requests so the agent can only request what data it actually needs.

Docs

There wasn't a whole lot of docs previously and so it could be a bit daunting to set it up and configure it. I've really tried to make it easier with a new docs section and installation instructions.

I'm keen for feedback

Please reach out to me if you use the tool, let me know what's working for you. Open a GitHub issue if you spot any problems or have suggestions for improvements.

Thank you to Anton

Thanks to Anton Tishchenko and his team for their hard work on the original v1 release, they did a monumental effort to get the coverage that this tool has, and I'm just standing on the shoulders of giants.

Anton took a lot of his personal time to review all my changes to the MCP server - thank you.