> For the complete documentation index, see [llms.txt](https://help.smartico.ai/welcome/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://help.smartico.ai/welcome/technical-guides/mcp-server/mcp-safe-rollout.md).

# MCP - Safe rollout

A short plan for giving your team access to the [MCP Server](https://help.smartico.ai/welcome/technical-guides/mcp-server) step by step: start with reading only, widen access when people are ready, and keep an eye on what each connection does.

## 1. Start everyone on a view-only profile

* Give new MCP users **MCP Viewer**. They can read records, read campaign journeys and search the product knowledge base. They cannot change anything, and they use no AI tokens.
* New back-office users start on **MCP No Access** until an administrator chooses a profile.
* Set the profile under **Label → BO users**, **General** tab, **MCP access profile**.

See [MCP - Access Profiles](https://help.smartico.ai/welcome/technical-guides/mcp-server/mcp-access-profiles.md).

## 2. Widen access step by step

| Step                 | Profile             | Adds                                                                          |
| -------------------- | ------------------- | ----------------------------------------------------------------------------- |
| Reading and learning | **MCP Viewer**      | Read only                                                                     |
| Data questions       | **MCP Analyst**     | Data warehouse SQL and golden queries (these use AI tokens)                   |
| Building             | **MCP Editor**      | Create and update, all operations (campaigns, flows, players), golden queries |
| Everything           | **MCP Full access** | All tools                                                                     |

* Need something in between? Create a custom profile with grants per tool and per operation, for example Editor without `stop_campaign_for_all`, or Viewer plus golden queries.
* When a person's profile changes, their personal sign-in is ended and they sign in again with the new profile.

## 3. One token per person or integration

* People should use **Sign in with Smartico**. Changes are recorded under their own name, and there is no secret to store.
* Create a token only for a script, an integration or a client that cannot sign in. Create **one token per person or integration**, with a clear name, the narrowest profile that works, and only the labels it needs.
* A token is shown in full only once. Store it like a password. Never share one token between people or tools.
* Changes made with a token are recorded under the label's AI/API user. A clear token name tells you which integration made them.

## 4. Ask for previews

Teach the team to ask the assistant to show its plan before every change ("show me what you will change first"). An optional dry run checks a change without saving it and answers "ok" or what is wrong. An update answers with the fields it changed, before and after. Actions that are not read-only, such as starting a campaign, make the AI client ask the user to confirm. New campaigns, missions and store items are created as drafts.

## 5. Review MCP → Usage

Every MCP call is listed under **Label → MCP → Usage**. Filter it by tool, entity type and token. Check it regularly, for example once a week in the first month:

* Who changed what, and with which token.
* Calls that failed or were refused: they often show that someone needs a different profile, or a better prompt.
* Tokens that are never used.

## 6. Deactivate what is not used

* Deactivate tokens that are not used under **Label → MCP → Access tokens**. Access ends at once.
* Deactivate the token of an integration that is retired, and the token of a person who left. A token belongs to the label's AI/API user, not to the person who created it, so deactivating that person does not stop the token.
* Give a person a narrower profile, or **MCP No Access**, when they no longer need MCP. Deactivating a person's personal token alone does not stop them: they can sign in again and get a new token while their profile allows it.
* The token list marks with a warning icon a service token whose creator can no longer log in. The token still works: check who uses it, and deactivate it if nobody needs it.

## 7. Mind the IP allowlist

If your account uses a back-office IP allowlist, it applies to MCP too. Add only the addresses you need, such as your office, your servers, and the addresses of ChatGPT or other web connectors your team uses. Claude's web connectors are accepted without it. See [MCP - Connect Claude, ChatGPT, Cursor and other MCP Clients](https://help.smartico.ai/welcome/technical-guides/mcp-server/mcp-connect-claude-chatgpt-cursor-and-other-mcp-clients).

## Checklist

* [ ] New users start on **MCP Viewer** (or stay on **MCP No Access**).
* [ ] Wider profiles only for people who need them.
* [ ] People sign in with Smartico; tokens only for integrations.
* [ ] One named token per person or integration, with the narrowest profile and labels.
* [ ] Previews before every change.
* [ ] **MCP → Usage** reviewed regularly.
* [ ] Unused tokens deactivated.


---

# Agent Instructions
This documentation is published with GitBook. GitBook is the documentation platform designed so that both humans and AI agents can read, navigate, and reason over technical content effectively. Learn more at gitbook.com.

## Querying This Documentation
If you need additional information that is not directly available in this page, you can query the documentation by asking a question.

Perform an HTTP GET request on the following URL with the `ask` and `goal` query parameters:

```
GET https://help.smartico.ai/welcome/technical-guides/mcp-server/mcp-safe-rollout.md?ask=<question>&goal=<user_goal>
```

`ask` is the immediate question: it should be specific, self-contained, and written in natural language.
`goal` is what the user is ultimately trying to achieve, the reason they need the answer. Sharing it helps GitBook give you a better, more relevant answer. A goal is most helpful when it describes the outcome the user wants rather than restating the question. For example, with `ask=how do I create an API token`, a goal like `automate deployments from our CI pipeline` lets GitBook tailor the answer to that use case.

The response will contain a direct answer to the question and relevant excerpts and sources from the documentation.

Use this mechanism when the answer is not explicitly present in the current page, you need clarification or additional context, or you want to retrieve related documentation sections.
