Same engine, different surface
Chats and the MCP server run on the same underlying tools against the same workspace data. What changes is where the conversation happens, what context Conversion provides, and how a few things behave once you’re connected:| Chats | MCP server | |
|---|---|---|
| Where it runs | Inside Conversion, on any page | Your own AI client (Claude, ChatGPT, Cursor, …) |
| Who drives the model | Conversion | Your client |
| Reading and lookups | Same tools | Same tools |
| Creating new assets | Staged as artifacts you review and apply | Direct create tools, live but dormant |
| Editing existing assets | Artifacts and in-product approvals | Raw, immediate edit tools |
| Confirmation before writes | In-product Allow / Deny | Handled by your client |
| Business context | Universal Business Context applied automatically | Not provided; your client is the model |
| Files and sandbox | Available | Not available |
Scope of a connection
Each connection is scoped to a single business and a single user:- You sign in with your Conversion account and authorize one of your businesses.
- The client can only see and act on data in that business, and only within your role and permissions there.
- To work with a different business, you connect again and authorize that one.
Next steps
Connecting
Add the server to your client, sign in, and choose a business.
Tools
Everything the MCP server can read, create, and edit.
Differences from chat
What’s the same as Chats, and what works differently.
Conversion AI
How Conversion AI works in the in-product chat.