AI assistants

Hypervalent Studio runs a hosted Model Context Protocol (MCP) server at https://mcp.hypervalent.io. Any MCP client — Claude Desktop, Claude Code, Cursor and others — can connect to it with an API token and work with your content the way a member does through the public API: read the content model, search and read documents, and, with a read + write token, draft, edit and publish.

Nothing runs on your side. The server is stateless and holds no data of its own: every tool call carries your token and is forwarded to the public content API with exactly that token's permissions.

Plan

AI assistants are part of the Team plan and above. On other plans the Use with AI panel in Studio shows an upgrade card, and the server answers every tool with an upgrade notice (FEATURE_DISABLED, requiredPlan: "team").

Where to start

I want to…	Read
Connect a client in five minutes	Quick start
Set up a specific client	Claude Desktop · Claude Code · Cursor · Other clients
Decide which token to give the assistant	Tokens & permissions
Have the assistant write content well	Authoring content
Use the built-in prompts, or borrow some	Prompts & recipes
Understand an error the assistant reported	Errors & limits
Fix a connection that will not work	Troubleshooting
See every tool and argument	MCP tool reference

How it works

Claude Desktop / Claude Code / Cursor
        │  MCP over HTTPS, Authorization: Bearer atk_…
        ▼
https://mcp.hypervalent.io/mcp?org=<org id>
        │  the same token, forwarded verbatim
        ▼
https://api.hypervalent.io/v1/orgs/<org id>/…   (the public content API)

1. Your client sends each tool call to the MCP server with your API token.
2. The server checks the token against the API once per call (a successful check is reused for 30 seconds), confirms the organization's plan includes AI assistants, and works out which project the call is about.
3. It translates the tool call into one or more public API requests, sent with your token.
4. The API applies the token's scope, project and rate limits, records any write in the organization's activity log attributed to the token, and answers.
5. The server returns the API's answer as JSON, with a plain-language hint on failures.

The server never sees the admin API: members, roles, billing and settings management cannot be reached through it, whatever the token.

What the assistant can do

Reads (any token)	Writes (read + write token, Team plan)
List the projects the token can reach	Create documents and taxonomy terms as drafts, fields and content blocks
List document types and read one as an authoring schema	Edit the working version of a document or term
List and read documents, published or working view	Publish a document — live immediately
Search titles, slugs and descriptions	Unpublish a document — off the public API immediately
List taxonomies, read a taxonomy and its terms	Trash a document (restorable in Studio)
Read General Settings and design tokens	Upload an image from a URL or from data

Two prompts ship with the server — draft-document and content-audit — and two resources expose a document type's schema and a document's content by URI. Every tool, prompt and resource with its arguments is in the reference.

What it cannot do

• Exceed the token. A project-narrowed token stays in its project. A read-only token lists every tool but is refused on every write. A revoked token fails on its next call.
• Manage the organization. No members, roles, billing, API tokens or settings changes.
• Change structure. Atoms, molecules, document types and taxonomies are edited in Studio.
• Restore from the trash or manage media beyond uploading.
• Search server-side. search_documents walks the document list page by page; keep searches narrow on large projects.

Safety in one paragraph

Drafts are private until published. Publishing, unpublishing and trashing take effect on the public API — and on every site reading it — at once, and the tools are annotated so clients that honour annotations ask first. Nothing the assistant can do deletes content permanently. Every write is attributed to the token in Studio → Activity, and revoking the token cuts the assistant off on its next call. Details in Tokens & permissions.