The modelling work that normally means clicking through a UI becomes something an assistant can carry out from a diagram, and the patch guide is a deliberate speed bump because patch operations have a different shape per entity. What deserves attention before you wire it up is the other half: this key can delete types, overwrite variants and unpublish live content, and the maintainers say plainly that content written by an editor can carry instructions a model will follow. Their own first recommendation is the one that holds regardless of model behaviour — give it a least-privilege key, so a hijacked call fails at the API rather than at the prompt.
A client for the Kontent.ai Management API covering the whole environment — content model, items and their language variants, assets, languages, collections, spaces, workflows and publishing. Every tool carries an MCP annotation marking it read-only, additive or destructive.
- The content model under full control: content types and reusable content type snippets listed, read, created, patched with move, addInto, remove and replace operations, and deleted (list-content-types, create-content-type, patch-content-type, delete-content-type, create-content-type-snippet, patch-content-type-snippet).
- Taxonomy groups on the same footing — listed, read, created, patched and deleted (list-taxonomy-groups, create-taxonomy-group, patch-taxonomy-group, delete-taxonomy-group).
- Content items and their language variants written as well as read: the item container, then a variant per language, updated element by element so omitted elements stay untouched, with a new version creatable from published content (create-content-item, update-content-item, create-content-item-variant, update-content-item-variant, create-new-content-item-variant-version, delete-content-item-variant).
- Reading that distinguishes draft from live: the current version of a variant, the published version specifically when a newer draft exists, and every translation of one item (get-content-item-variant, get-published-content-item-variant-version, get-content-item-translations).
- Finding content two ways — filtered and searched listing of variants, a bulk fetch of specific item-and-language pairs with a continuation token, and semantic search by meaning for when you do not know the keywords (list-content-item-variants, bulk-get-content-item-variants, search-content-item-variants).
- The publishing lifecycle: move a variant to another workflow step, publish or unpublish it now or scheduled to a future date and timezone, and cancel a scheduled publish to reopen it for edits (change-content-item-variant-workflow-step, publish-content-item-variant, unpublish-content-item-variant, cancel-scheduled-publishing-content-item-variant).
- Workflows themselves — created with custom steps, transitions, scopes and role permissions, updated, and deleted when nothing uses them (list-workflows, create-workflow, update-workflow, delete-workflow).
- The environment's structure: assets and asset folders, languages, collections that bound content by team or brand, spaces for a website or channel, and the role list (list-assets, update-asset, patch-asset-folders, list-languages, create-language, patch-collections, create-space, list-roles).
- A per-entity patch guide the server insists you fetch before attempting any patch operation (get-patch-guide).
A Kontent.ai account with a project, a Management API key, and the environment ID. Run `npx @kontent-ai/mcp-server@latest stdio` with KONTENT_API_KEY and KONTENT_ENVIRONMENT_ID for a local client, or `shttp` to serve several environments from one instance on port 3001, where each request carries its environment in the URL path and its key as a Bearer token.
One command plus a key — npx @kontent-ai/mcp-server@latest stdio, then supply credentials