Daktela MCP - AI server¶
Connect your AI assistant (Claude, Copilot, Cursor and others) to the Daktela AI platform. The assistant can then build, edit, test and analyze your chatbots, voicebots and emailbots directly from the conversation.
How it works¶
The server is hosted by Daktela, so there is nothing to install. You add one address to your AI client and sign in once at the start of each conversation, choosing the instance the assistant may work with and the permission level it gets.
After that you describe what you need in plain language. The assistant picks the right tools itself and follows the platform's conventions for dialog design.
Tip
This is the server for the Daktela AI platform. For an assistant that works with your Contact Centre data instead (tickets, calls, CRM...) see Daktela MCP - CC.
What it can do¶
| Area | What the assistant does |
|---|---|
| Overview | Reads your dialogs, scenarios, AI agents, integrations, goals and instance settings. |
| Build | Builds a full dialog from a description, or adds single nodes: messages, buttons, forms, conditions, AI agents, integrations, lambdas, redirects, end call, emails. |
| Safe edits | Works on the TEST version of a dialog, never on the one your customers reach. |
| Testing | Writes and runs automated tests, and holds live test conversations with your bot. |
| Traffic analysis | Filters production discussions, reads transcripts, finds discussions by feedback tag, shows the debug trace for a single discussion, adds tags and comments. |
| AI knowledge base | Manages sources, URL lists and manual entries. Checks what the knowledge base finds for a given question. |
There are two permission levels, Read only and Read & Write. You pick one when you sign in, and the assistant cannot go beyond it.
How edits stay safe¶
Every dialog can hold several versions, but only one is Production and only one is Test; the rest are drafts. The assistant works on the Test version, so the dialog your customers reach keeps running untouched. You test the changes, and they only go live when you tell the assistant to publish, or publish them yourself in the administration.
See Dialogs β Versioning for how versions work.
Example requests¶
New bot from a description¶
Build a voicebot for complaints. It checks the order number through our API and asks for the reason. If the customer wants a person, it transfers the call to an operator. Then write tests for it.
Safe change to a live dialog¶
The assistant makes the change on the Test version, tests it, and publishes only when you say so.
In the Delivery dialog, add a branch for parcel locker questions. Test it and show me what changed. Do not publish yet.
Regression tests before publishing¶
Run all tests of the Orders dialog. For each test that fails, tell me why.
Talk to the bot as an angry customer who wants to cancel the contract. Tell me where the bot gets stuck.
Debugging a single discussion¶
The assistant reads the debug trace: the scenarios the conversation went through, how context variables changed, and what the integrations returned.
Discussion 184223 went to an operator, but the customer only wanted the order status. Find out why.
Production traffic analysis¶
Look at last week's discussions that ended without a completed goal. Summarize the most frequent causes.
Find the discussions that supervisors tagged as a wrong answer, and suggest changes to the prompt.
Knowledge base maintenance¶
Does the knowledge base find an answer to 'how long does a refund take'? If not, add a manual entry from this text.
What's Next?¶
Ready to connect your AI client? See Set Up Your AI Assistant.