What's new in Core dna MCP 2.0: fewer tools, a lot more capabilities

What our CTO built from one brief

Here's what building on Core dna looks like at full scale. To test the release, our CTO, Dmitry Kruglov, gave Claude a brief for one of our clients. They collect a lot of emails through web forms, some with more than 300,000 submissions, and had no reporting on any of it.

The brief set out the outcome (a live report page for each form), the approach (one automation checks for new submissions and hands the work to another) and the constraints (no endless loops, big backlogs handled in batches, no two runs writing at once). Claude wrote a plan, Dmitry reviewed it and answered its questions, and then it built, on a development copy of the client's site:

• A new kind of content that stores the report data and shows each report as a page with a private, hard-to-guess address
• An automation with 13 steps that checks each form for new submissions
• A second automation with 26 steps that reads the new submissions and updates the numbers

The screenshots below look technical, and you never need to build or read anything like them yourself. Claude built every step from the brief. You describe what you want in plain language, and the AI builds it for you.

To turn on reporting for a form, someone on the team creates a report and connects it to that form. "Now we can build full-blown apps on Core dna from Claude," Dmitry told the team afterwards.

Before 2.0, a build like this would have been a development project. Now it starts with a brief.

Build automations by describing them

Say what should happen and when, and the agent builds it: an email to your sales team every time someone fills in the contact form, a sale that starts on Black Friday and ends after Cyber Monday. An automation can start from a form, a schedule, a change to any record or a webhook, and it can branch and repeat.

If something goes wrong, the agent runs the automation, finds the step that failed, explains it and runs it again. Your team checks the result, the same way you'd review a colleague's work. In the Core dna admin these are called jobs. See how automations and workflows run in Core dna.

Add new kinds of content

Events, case studies, product guides, store locations. Describe the content you need and the agent sets it up with the fields, web addresses, categories, tags and publishing options to match, then adds the first entries as drafts.

Put a new kind of content together with an automation and you have a working feature, built in one conversation. That's how the analytics app above was made.

Change one part of a page

Ask for one paragraph to change and only that paragraph changes. Everything else on the page, post or FAQ stays exactly as it was, and the previous version is kept, so every change is easy to check and easy to undo.

It asks before anything risky

• Deletes need your OK. If removing something would take other content with it, the agent has to confirm first, and it tells you exactly what would go.
• No passwords. The MCP never sets passwords. New users set their own.
• Safe uploads. Files can only be added from public, secure web addresses.
• Hidden credentials. Keys and passwords inside automations are masked.

Also improved

Smaller changes that add up when you work on your site from chat:

• Tag by name. Say "tag this as Product News" and it's done. If two tags sound alike, the agent asks which one you mean.
• Find any image or file. Search by name, ALT text, description or tag, and see why each one matched.
• Find a page from its web address.
• See your full menus, with the page behind every link.
• Read forms and their entries by field name.
• Every site and language in your account in one view, so the agent works on the right one.
• Clearer messages when something can't be done, so the agent can correct course without guessing.