Allowing and Approving Changes by your Agent
On This Page
The Hevo MCP server is read-only by default. Your agent can view your Pipelines, Destinations, and other resources, but cannot change them. To let the agent make changes, you set a scope, which is the level of changes that the agent can make. Within that scope, the server asks you to approve each risky change before it sends the change to Hevo.
The scope and the approval only limit what the agent can do. They never allow the agent to perform an action that your role does not allow.
Scopes
The following table lists the actions that your agent can perform with each scope. The write scope includes the actions of the read scope, and the delete scope includes the actions of both:
| Scope | The agent can |
|---|---|
read |
View your Pipelines, Destinations, Models, Workflows, team users, and OAuth accounts. This is the default scope. |
write |
- Create and update Pipelines, Destinations, and Models. - Enable, disable, pause, resume, run, sync, and resync Pipelines. - Cancel the jobs of a Pipeline. - Include, exclude, pause, resume, skip, or restart objects, and include or skip Event Types. - Change the schedule, priority, position, query mode, schema mapping, Auto Mapping, Source configuration, or transformation code of a Pipeline. - Pause, resume, run, and reset Models, and change their schedule. - Load data into Destinations now. - Run Workflows. - Invite team users or change their roles. |
delete |
Delete Pipelines, Edge Destinations, team users, and OAuth accounts. |
If the agent tries to make a change that its scope does not allow, the server rejects the request without sending it to Hevo. The agent then tells you that the change is not allowed and how to allow it.
After each change, the agent reads the resource again and reports its new state. Some actions, such as disabling a Pipeline, take time to complete after Hevo accepts the request, so the agent reads the Pipeline again to check its state.
Setting the Scope
The HEVO_MCP_DEFAULT_SCOPE environment variable sets the scope. The following table lists the steps to set the scope for each agent:
| Agent | Steps |
|---|---|
| Claude Code or Codex CLI | - Add export HEVO_MCP_DEFAULT_SCOPE=write (or delete) to your shell profile file, such as ~/.zshrc. On Windows, run the setx HEVO_MCP_DEFAULT_SCOPE write (or delete) command in PowerShell instead.- Open a new terminal. - Start a new session of your agent from that terminal. |
| Codex app | - Add export HEVO_MCP_DEFAULT_SCOPE=write (or delete) to your shell profile file, such as ~/.zshrc. On Windows, run the setx HEVO_MCP_DEFAULT_SCOPE write (or delete) command in PowerShell instead.- Quit the Codex app, and open it again. |
| Claude Desktop | - Add "env": { "HEVO_MCP_DEFAULT_SCOPE": "write" } (or "delete") to the hevo entry in the claude_desktop_config.json file.- Quit Claude Desktop, and open it again. |
After you restart your agent, the hevo_whoami tool reports the new scope.
Approving Changes
Some changes need your approval every time, in every scope. Before the server sends such a change to Hevo, it shows you an approval prompt in your agent with the following details, in this order:
- The change, the resource, and the region, such as Delete Edge Destination 352 in the us region?
- The effect of the change. For example, the prompt to delete a Destination states that the deletion also deletes every Pipeline that loads data into the Destination, and that it cannot be undone.
- A reminder that your agent cannot approve the change for you.
- The technical details of the request, which are the endpoint and its path, the query, the connection fields of a Source or a Destination, such as its host, and the request body. The prompt never shows a password. It shows a reference to a secret file, or
[redacted], in its place. The prompt shortens a long request body to its first 1,500 characters, unless the request contains a reference to a secret file. - The name of the secret file and the field in which Hevo uses its content, for a request that contains a reference to a secret file. If the request contains the host of the Source or Destination, the prompt shows it on its own line. Check that this is the host of your Source or Destination.
The agent cannot answer the prompt for you, and no scope or setting skips it.
When you ask for the same change to many resources, such as pausing 40 objects of a Standard Pipeline or disabling 40 Edge Pipelines, the server shows one prompt for each group of up to 100 resources. The first line of the prompt states the number of resources. The prompt then lists the first 60 resources and the number of remaining resources. After you approve it, the server makes the change for each resource and reports the result of each. The approval applies only to these resources. If a change fails because of an authentication, permission, or rate limit error, or because of a connection error or a timeout, the server skips the remaining resources and reports them as skipped. If the change fails for one resource for any other reason, such as a resource that does not exist, the server reports the error for that resource and continues with the next one.
In Claude Code, the prompt shows the first line and the effect of the change. The reminder and the technical details are collapsed into a line such as … (+7 more lines), which cannot be expanded. To see the request before you approve it, press Ctrl+O to open the transcript, which shows the arguments of the call, such as the ID and the request body. Press Ctrl+O again to return to the prompt.
In the Codex CLI and the Codex app, the prompt shows all the details. The Approve for me setting of the Codex app does not answer the approval prompt for you.
Before the prompt, Codex can ask for your permission to run the tool, depending on its approval setting. This permission is separate from the approval prompt of the server.
The following changes need your approval:
-
Every delete. Before the agent deletes a Standard Pipeline, it asks you whether to also delete its tables in the Destination.
-
Configuration changes to a Destination, a Source, a Model, or a Pipeline and its objects, such as the schedule, the priority, the position, the query mode, the schema mapping, Auto Mapping, or the transformation code.
-
Including or excluding an object, or including an Event Type.
-
Actions that stop or re-ingest data, such as canceling a job, disabling or pausing a Pipeline, pausing a Model, pausing or skipping an object, skipping an Event Type, resyncing, restarting, or resetting a Model.
-
Runs, such as running a Pipeline now, syncing now, loading a Destination now, or running a Model or a Workflow. The data that a Pipeline run or sync ingests counts toward your Events quota.
-
Enabling an Edge Pipeline, and resuming an object.
-
Inviting a user, and changing the roles of a user.
-
Any request that contains a reference to a secret file, such as creating a Destination with its password in a secret file.
The prompt contains an Approve or Reject choice, with Approve selected by default. To approve the change, keep Approve selected and click Accept. In Claude Code, press the Down arrow key to move to Accept, and press Enter. After you approve the change, the server sends it to Hevo, and the agent reports the new state of the resource.
To reject the change, select Reject and click Accept, or click Decline, which is Deny in the Codex app. In Claude Code, press Esc to cancel the prompt.
In the Codex CLI, use the arrow keys to select Approve or Reject, and press Enter. Press Esc to cancel the prompt.
The server sends nothing to Hevo in the following cases:
-
You reject the change or cancel the prompt: The agent does not retry the change, or look for another way to make it, unless you ask again.
-
Nobody answers the prompt within 10 minutes: The prompt expires. Depending on your agent, ask the agent again, or answer the new prompt that the server shows.
-
Your agent cannot show approval prompts: Claude Desktop cannot show approval prompts. The agent tells you that it cannot request your approval. Make the change in the Hevo UI, or use Claude Code in a terminal or Codex.
-
Your agent rejects the prompt without showing it: Claude Code in print mode and in its Visual Studio Code extension rejects every approval prompt without showing it to you. Make the change from Claude Code in a terminal, or in the Hevo UI.
-
The agent runs without a user: In a scheduled or scripted run, nobody can answer the prompt, so every change that needs approval is canceled. Use scheduled runs only for tasks that do not need approval, such as a daily health report.
Approval Prompts in each Agent
The following table lists whether each agent can show approval prompts:
| Agent | Approval prompts |
|---|---|
| Claude Code in a terminal | Supported. |
| Claude Code in print mode or in its Visual Studio Code extension | Not supported. The agent rejects every prompt without showing it. |
| Codex CLI and Codex app | Supported. |
| Claude Desktop | Not supported. Changes that need approval are not sent to Hevo. |
An agent that cannot show approval prompts can view your resources and make the changes that do not need approval.
Changes that Do Not Need Approval
The following changes do not need your approval:
-
Creating a Pipeline, a Destination, or a Model, unless the request contains a reference to a secret file.
-
Resuming a Standard Pipeline or a Model.
The server sends these changes to Hevo without an approval prompt. The hevo-pipelines skill instructs the agent to make such a change only when you ask for it, and to show you the request body before it sends the request. For example, a general request such as Fix my Pipelines is not a request to create anything.
Reading your resources never needs approval.
Revision History
Refer to the following table for the list of key updates made to this page:
| Date | Release | Description of Change |
|---|---|---|
| Nov-18-2026 | 2.54 | New document. |