> For clean Markdown of any page, append .md to the page URL.
> For a complete documentation index, see https://docs.ada.cx/llms.txt.
> For AI client integration (Claude Code, Cursor, etc.), connect to the MCP server at https://docs.ada.cx/_mcp/server.

# MCP: Change sets — draft, test, promote, and revert Agent changes

Behavior changes made through the MCP server no longer apply to live conversations as soon as they are saved. They are staged on a change set, can be tested against simulations first, and go live only after explicit approval. A change set that has gone live can be reverted in one step.

## What's new

Change sets add a draft, test, promote, and revert loop to Agent behavior changes.

* **Staged edits** — [`edit_agent_behavior`](/mcp/tools/edit-agent-behavior) bundles changes to knowledge articles, coaching, Playbooks, Actions, and custom instructions on a change set that does not affect live conversations. Up to 10 entities per change set, and 5 active change sets per Agent.
* **Test before promoting** — run test cases against a change set and the simulated conversation uses the staged edits instead of the live ones. Change sets are isolated from each other, so different versions of the same entity can be tested independently.
* **Approval before anything goes live** — promoting, reverting, or discarding a change set requires explicit confirmation, and cannot be skipped.
* **Conflict warnings** — when it comes time to promote, your MCP assistant flags any of your edits that conflict with a newer change on the live Agent, any entity you have edited that no longer exists, and any edit that references an entity that no longer resolves. Conflicts are surfaced rather than merged.
* **All-or-nothing promotion and clean revert** — every edit in a change set is applied together. A promoted change set can be reverted to the state captured when it was promoted.
* **Change set visibility** — [`list_agent_changesets`](/mcp/tools/list-agent-changesets) lists change sets by status and returns a before-and-after comparison for any one of them.

## Changes to existing behavior

Two changes affect how existing MCP tools are used.

* `propose_change` is now [`edit_agent_config`](/mcp/tools/edit-agent-config) and covers settings such as test cases, topics, custom metrics, scorecards, and Custom Glossary terms. Its changes continue to apply immediately.
* Knowledge articles, coaching, Playbooks, Actions, and custom instructions are edited with [`edit_agent_behavior`](/mcp/tools/edit-agent-behavior) instead, and are staged rather than applied immediately.

## Availability

Change sets are managed through the MCP server. There is no change sets screen in the Ada dashboard, and test runs pinned to a change set do not appear on the Simulations page — retrieve their results with [`get_test_runs`](/mcp/tools/get-test-runs).

Write tools require an Owner or Admin role when using OAuth. See [Permissions](/mcp/introduction/authentication#permissions).