# MCP conversations

> Each conversation an AI assistant has with Nodaro over MCP becomes a workflow of connected nodes that you can open in the editor, change and run again.

Source: https://nodaro.ai/docs/mcp/conversation-flows

When an AI assistant makes something through the [Nodaro MCP server](https://nodaro.ai/docs/mcp), each conversation becomes a **workflow** in your Nodaro account. Every image or video the assistant generates is a node, and when one result feeds the next step, the two nodes are connected. You can open the workflow in the editor, see what was made and how, change a setting on one node and run that node again.

This works for Claude, ChatGPT, Claude Code and the other clients that connect to Nodaro. With every step, the assistant also gets a link to the workflow, so it can send you the link to open in the editor.

## Which steps become nodes

These tools add a node to the conversation's workflow:

| Tool | Node |
| --- | --- |
| [`generate_image`](https://nodaro.ai/docs/mcp/tools/image#generate_image) | [Generate Image](https://nodaro.ai/docs/nodes/image/generate-image) |
| [`modify_image`](https://nodaro.ai/docs/mcp/tools/image#modify_image), [`image_to_image`](https://nodaro.ai/docs/mcp/tools/image#image_to_image) | [Modify Image](https://nodaro.ai/docs/nodes/image/modify-image) |
| [`animate_image`](https://nodaro.ai/docs/mcp/tools/video#animate_image), [`generate_video`](https://nodaro.ai/docs/mcp/tools/video#generate_video) | [Generate Video](https://nodaro.ai/docs/nodes/video/generate-video) |

A call that a node cannot repeat exactly does not become a node. Examples are several images in one call, with `count` or with more than one model, and a palette or moodboard direction. The result of such a call is still saved in your library and listed under **Recent from MCP**.

An image that the workflow did not make, such as an upload or an image from another conversation, appears as an [Upload Image](https://nodaro.ai/docs/nodes/image/upload-image) node. That node is connected to the step that used the image.

## One conversation, one workflow

Nodaro keeps one workflow per conversation:

- **ChatGPT** and **Claude Code** tell Nodaro which conversation a request belongs to. Each chat gets its own workflow. For Claude Code, each launch is one conversation.
- **Claude** on the web and in the apps does not tell Nodaro. Its steps join the same workflow until there is a pause of more than **30 minutes**, and the next step then starts a new workflow. Two Claude chats that run side by side within 30 minutes share one workflow.

Each workflow is named after your first request in the conversation, shortened to about 50 characters. When the request is not available, the workflow is named after the assistant's first prompt.

### Your own words

The tools above take an optional `user_request` argument. The assistant is asked to pass your latest message, word for word. Nodaro keeps it on the node, under **Where it came from**, and uses it to name the workflow. It never changes what is generated.

Not every assistant passes it. An assistant that keeps an old copy of Nodaro's tool list does not see the argument until its connector refreshes the list.

## Where the conversations are

Open the home screen and choose **MCP Conversations** in the **Jump back in** row.

- **Month and year.** The tab shows one month at a time. The arrows change the year. The month menu marks the current month and shows how many conversations each month has, and a past month with none cannot be chosen. A conversation belongs to the month it started in.
- **Source.** Keep only some assistants: Claude, Claude Code, ChatGPT, Codex, Cursor and any other client you connected. The connections of one product count as one source, so "Claude Code (nodaro)" and "Claude Code (staging)" are one Claude Code. Hover over a source to see its connections one by one.
- **Each card** shows these details:
  - The assistant, and when the conversation started.
  - The conversation's name, and how many nodes it has of each kind.
  - The credits it used.
  - A cover: the one you set in the editor, or else its latest result.
  - A highlight while the conversation is still receiving steps.
- **The card menu** has these items:
  - **Open**.
  - **Export JSON**, a template, as the editor exports it.
  - **Copy to project…**, a standalone copy in a project you choose. The conversation stays where it is, and the copy keeps each node's **Where it came from**.
  - **Rename**.
  - **Use the latest result as cover**, when you set a cover before.
  - **Delete**.

**Deleting a conversation** deletes its workflow. The images and videos the assistant made stay in your library and under **Recent from MCP**. The results of runs you started in the editor are deleted with the workflow, as for any workflow.

## In the editor

The canvas of a conversation says which assistant is building it, and links back to **MCP Conversations**.

- Each node an assistant made shows **Where it came from** first in its settings panel: your request and the assistant. Under **Details**, it shows the tool the assistant called and when.
- Steps the assistant adds while you have the workflow open appear on the canvas by themselves.
- When you have unsaved edits, the editor tells you that the workflow changed elsewhere, instead of replacing your work.

## From code

The conversations are listed programmatically too:

- REST: `GET /v1/mcp-conversations?month=YYYY-MM&sources=claude,chatgpt`. See [Workflows › MCP conversations](https://nodaro.ai/docs/developers/api/workflows#mcp-conversations).
- SDK: `client.workflows.listMcpConversations({ month, sources })`. See [SDK › Workflows](https://nodaro.ai/docs/developers/sdk/workflows#listmcpconversationsparams).
- CLI: `nodaro workflows mcp-conversations --month YYYY-MM --source claude`.

Each conversation is a workflow in your **mcp** project, filed under a hub workflow for its month. The hubs do not appear in your workflow lists.

## Turn it off

Use the **Turn my conversations into flows** switch on the **MCP Conversations** tab. The same setting is available through the API:

```http
PATCH /v1/user/settings
{ "mcpPreferences": { "flows": { "enabled": false } } }
```

With the setting off, MCP jobs run exactly as before and appear only in your library and under **Recent from MCP**. Turn the switch back on, or send `true`, to start again.

## Frequently asked questions

### What happens to the images and videos an AI assistant makes through Nodaro?

Each conversation becomes a workflow in your Nodaro account. Every image or video the assistant generates is a node, and a result that feeds the next step is connected to it. You can open the workflow in the editor, change a setting and run one node again.

### Where do I find my MCP conversations?

Open the home screen and choose MCP Conversations in the Jump back in row. The tab shows one month at a time, and you can keep only some assistants with the source filter.

### How does Nodaro know which steps belong to one conversation?

ChatGPT and Claude Code tell Nodaro which conversation a request belongs to, so each chat gets its own workflow. Claude on the web and in the apps does not, so its steps join one workflow until there is a pause of more than 30 minutes.

### Does deleting an MCP conversation delete what the assistant made?

No. Deleting a conversation deletes its workflow, but the images and videos the assistant made stay in your library and under Recent from MCP. Results of runs you started in the editor are deleted with the workflow.

### Can I stop my conversations from becoming workflows?

Yes. Turn off the Turn my conversations into flows switch on the MCP Conversations tab. MCP jobs then run as before and appear only in your library and under Recent from MCP.