Skip to main content
The MCP server and Conversion AI in chat share product tools, workspace feature gates, and permission checks. The available operations also depend on the session. Chat tool settings govern chat; MCP approvals are configured in your external client.

​
Approvals and confirmation

Your MCP client controls confirmation before tool calls. When it sends a write request, Conversion executes it subject to backend permissions and workspace feature gates. Chat’s disabled and approval-required tool settings do not hide MCP tools. Chat uses Conversion’s in-product approval flow. MCP uses the external client’s flow, so review that client’s approval settings before using write tools.

​
Editing existing assets

Chat uses edit_* tools. An edit configured for approval is added to the pending edit list and applied after approval; an edit configured for automatic execution applies immediately. MCP uses separate direct update tools, such as update_email, update_form, and update_campaign. They write to the relevant update endpoint when the client calls them, after any client-side confirmation. MCP does not use chat’s pending-edit flow or its internal execute_edit tool. Read the current asset before changing it. Full email bodies and form designs replace the existing definition. Template edits and replacement of existing workflow graphs are available in chat, but not over MCP.

​
Files and the sandbox

Both surfaces can work with hosted Files and Media items, subject to permissions. MCP can list and retrieve them, and can edit metadata or delete items when backend permissions allow it. Only chat has a Conversion sandbox for attachments, scripts, generated files, and uploads. MCP does not expose sandbox-dependent CSV workflows, image generation, or hosted browser tools. Your client’s local file handling does not provide access to Conversion’s sandbox.

​
Universal Business Context

Chat loads workspace and personal context for the model. MCP does not automatically provide that context to your external model. Add the business guidance and brand voice you want it to follow to your client’s instructions.

​
How skills load

Chat reads building references from its sandbox. MCP exposes matching learn_* tools for available capabilities. learn_reporting includes its example definitions, so an MCP client does not need a filesystem to read them. For product documentation, your client can browse Conversion Docs or connect the separate public documentation MCP server at https://docs.conversion.ai/mcp.

​
At a glance

The tool reference explains deletes, external actions, and the restrictions on Play and webinar operations.