The assistant and MCP
A scenario can be described in words
Describe who the bot answers and what it does when the answer is not in the documents. The assistant places the nodes, links them and shows a list of its edits: until you press Apply, nothing on the canvas changes.
Scenario assistant
If the answer is not in the documents, have it open a ticket in the CRM and tell the customer a manager will call back
Added a branch after the search and a «Ticket in the CRM» node on the «not found» side. Put the text for the customer into the answer.
What changed
- +Node «Found in the documents?»
- +Node «Ticket in the CRM»
- +3 links
- ~Edited the text in «Answer to the customer»
Validated the scenario and ran it dry: it starts
One sentence asked, and what the assistant proposes. The two new cards are marked on the canvas, and the panel lists what was added and what changed.
300 quarks a month once your address is confirmed, no card needed. Talking to the assistant spends the same quarks as answering customers; running a scenario dry costs nothing.
Once the bot is answering customers
«Will it redo everything its own way?»
The canvas stays yours
The assistant does not hide the result behind a dialogue: every edit lands on the canvas marked. Before showing it, the assistant runs the scenario through the same check that guards publishing, and runs it dry: no email leaves, no ticket is opened, the customer gets nothing.
- It shows what it did. A list of edits, not «done».
- It does not touch what is published. It works in the draft only.
- It edits what you asked about. What you built by hand is left alone.
- One button undoes it. That is a rollback of its edit, not an undo of your last action.
For people who already run an agent
Iterna as a tool inside your MCP client
If you work in Claude Code, Cursor or another MCP client, connect the workspace with one line and build scenarios from there.
The key is issued per workspace and grants read access only by default: writing, running and deleting are switched on one at a time, and what is not switched on never appears in the tool list. A run over MCP is an ordinary run: it counts against the workspace owner's quarks and shows up in the run list. How to connect it is written up in the documentation.
$ claude mcp add --transport http iterna-mcp \
https://iterna.ai/api/mcp \
--header "Authorization: Bearer amcp_…"- list_workflowsread
- get_executionread
- validate_workflowread
- kb_searchread
- patch_workflowwrite
- run_workflowrun
Describe your bot in one sentence
One sentence about who to answer and how is enough for a first scenario. After that it is visible on the canvas and edited by hand.
Describe your bot in words