Release Notes

Release Notes

Every Core dna product, release by release, newest first. Each product ships on its own schedule.

MCP Server 2.0.0

What’s new in Core dna MCP Server 2.0.0

Jobs, content types and fewer tools

Agents can build, run and debug complete jobs and create their own content types from chat, with 65 clearer tools in place of 81.

Fewer, clearer, more capable tools

The server now has 65 tools in place of 81, so agents find the right tool faster and use less context.

Build and run jobs from chat

Agents can now create, edit and delete complete jobs: triggers (record events, form submissions, webhooks), a schedule, and actions with conditions and loops. They can run a job, re-run a failed run, and explain a run action by action to find the step that failed. Jobs that start other jobs work, and credentials in job settings are masked.

Create your own content types

Agents can now create and edit custom entity types: name, identifier, fields through the layout, options such as URLs, categories, tags and publishing, and relationships to other types. A guard stops relationships from being dropped by accident. Together with jobs, an agent can build a working feature from chat: the data model, the automation behind it, and the records.

One tool for revisions

One edit tool now makes partial changes to any entity with revisions: pages, blog posts, FAQs, books and custom entity records. It updates an area, patches, inserts, appends or removes components in a collection, and changes the revision name or layout, without sending the whole revision again.

Categories and tags made simple

One tool lists categories, tags, layout categories, job categories, countries and states, and another creates and edits the first four. When agents tag a page or a blog post, they can use the category or tag name. The server finds the id, and if a name is ambiguous, it returns the candidates.

Find any file or image in one call

Search checks the file name, ALT text, description, tags and categories together, and says why each file matched. An upload takes a folder path, ALT text, tags and categories in the same step. One tool renames, moves or re-describes an asset.

Safer by default

Deleting a file, a folder with files, or a content type with records now needs an explicit confirmation, and the answer names what is deleted with it. The server never sets passwords: new users set their own through a reset or a passkey. Uploads from a URL accept only public HTTPS addresses.

Other improvements and bug fixes

  • Tool descriptions are about a quarter smaller: about 58k tokens for the full set, down from about 76k.
  • JSON answers are sent compact: about 30% fewer tokens for the same data.
  • One call returns every site with its hostnames, homepage, languages, currencies and timezone, so agents use the correct site and language.
  • Site data: agents read and write the key-value store per site or account-wide. Values are masked unless an agent asks for them.
  • Menus are returned as full trees, with the name and path of each linked page, and agents can add items by menu name.
  • A page can be found by its URL; "/" returns the homepage.
  • Forms include their fields, and submissions are keyed by field name.
  • Products can include their variants, and books their units and pages.
  • Layouts can be listed for a record type, and a layout category set by name.
  • Clearer errors: a bad sort field is refused with the list of valid fields, temporary server errors say so, and when a write may have applied, the agent is told to check before it retries.
  • New users must have the frontend or backend type and be linked to a person.
Go deeper

Was this page helpful?

Your answer goes to the docs team.