> For the complete documentation index, see [llms.txt](https://integrations.drivetrain.ai/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://integrations.drivetrain.ai/extending-drivetrain/using-the-drivetrain-mcp-server.md).

# Using the Drivetrain MCP Server

Access Drivetrain metrics via AI clients using MCP.

The **Drivetrain MCP server** connects Drivetrain to **MCP-compatible AI clients** such as [ChatGPT](https://integrations.drivetrain.ai/extending-drivetrain/using-the-drivetrain-mcp-server/connect-common-mcp-clients.md#connecting-to-chatgpt) and [Claude](https://integrations.drivetrain.ai/extending-drivetrain/using-the-drivetrain-mcp-server/connect-common-mcp-clients.md#connecting-to-claude-oauth), so you can ask for your metrics in natural language from that client.

### What is an MCP server?

An **MCP (Model Context Protocol) server** exposes **tools**: named actions an AI client can call. The client decides which tool fits your prompt, calls it, and works with the result. Clients such as [Claude Desktop](https://claude.ai/download) can then compute derived metrics, chart results, or combine Drivetrain data with other MCP servers (for example, draft an email with a Gmail server, or update a Notion page).

{% embed url="<https://drivetrain-ai.wistia.com/medias/pte5hcgkf1>" %}

### Scope of the Drivetrain MCP Server

The server exposes up to 22 tools. Which ones you see depends on the **MCP** setting on your role (see [Set up MCP access](https://integrations.drivetrain.ai/extending-drivetrain/using-the-drivetrain-mcp-server/set-up-mcp-access.md)).

* **Read tools (18)** look up metrics and their values across time, dimensions, versions, and scenarios, plus dimensions, Lists, connected datasets, boards, and your model's structure. They never change anything.
* **Write tools (4)** create or update a board, or add or change a dataset transformation. Each saves a **draft**; nothing takes effect until you open the draft in Drivetrain and save it.

The server can't edit your model, change values, or delete anything you've saved; write tools only create or update drafts. See [Drivetrain MCP tools](https://integrations.drivetrain.ai/extending-drivetrain/using-the-drivetrain-mcp-server/available-tools.md) for every tool, its inputs, and its permission rules.

### Example

Prompt: <mark style="color:$primary;">`What's my Revenue per FTE?`</mark>

1. The client searches your metrics for **Revenue per FTE**.
2. If it exists, the client fetches its values. If not, it fetches **Revenue** and **FTE** and divides them itself.
3. You get the result for the period you asked about, for example month by month across 2025.

Follow-ups such as benchmarking against industry data or drafting recommendations are done by the AI client with its own tools (for example, web search), not by Drivetrain.

### What respects your permissions

Every person connects with their own Drivetrain account, and every tool call runs as that person:

* **Metrics** you can't access are never returned, and are removed from any board a write tool saves.
* **Dimension values** outside your access are dropped when a tool looks them up, and metric values are calculated within your access. Dimension *names* aren't hidden: `list_dimensions` lists every dimension.
* **Scenarios**, **boards**, **Lists**, and **datasets** follow the same access as in the app. Some tools need more, such as admin or **Full access** (listed per tool in [Drivetrain MCP tools](https://integrations.drivetrain.ai/extending-drivetrain/using-the-drivetrain-mcp-server/available-tools.md)).
* The role check runs on every call, so a role change applies to the very next call.

### Limits and behaviour

* `query_metrics` takes at most **50 metrics** and **5 dimensions** per call. Dimensions must apply to every metric in the call.
* When no scenario is given, `query_metrics` uses **Base Case** (an Actual-only request has no scenario).
* If Drivetrain can't confirm your permissions within about 8 seconds, the call is refused rather than run unchecked. Retry it.
* Write tools never publish. A board or transformation draft waits in Drivetrain until you **Save** or **Discard** it.

## FAQs

<details>

<summary>Where is my data hosted and processed?</summary>

Drivetrain, including the MCP server, is hosted on AWS in the US East (N. Virginia) and EU (Frankfurt) regions, inside the same security perimeter as the rest of the product. EU accounts use an address ending in `.eu.drivetrain.ai`.

When you use Claude, Anthropic also processes your prompts and the data the tools return, under its API data policy. That data isn't used for model training. Where Anthropic processes it depends on how your organization's Claude account is set up.

</details>

<details>

<summary>Does the MCP connection get the same access as the user who connects it?</summary>

Yes. Each person signs in with their own Drivetrain account, so the connection has exactly that user's access. Drivetrain checks the user's role on the server for every tool call. There's no shared service account: the API key method also uses the admin's own personal key.

</details>

<details>

<summary>Can the MCP server return more data than a user is allowed to see?</summary>

No. It follows the same permissions set on the [**Roles**](https://github.com/DrivetrainAi/gitbook-integrations/tree/main/en/user-administration-and-access/access-control-and-permissions/creating-and-managing-roles.md) page, and each token is tied to one user and one account.

**Note:** Once data is returned to the AI client, it becomes part of that user's conversation. If you later reduce their role, data they already fetched stays in their chat history. If they share the conversation or a Claude project, the people they share it with see that data too.

</details>

<details>

<summary>Are MCP tool calls logged, and can I see the logs?</summary>

Yes. Drivetrain logs every tool call that reaches the MCP server, including ones a permission check refuses, with:

* Timestamp
* User and account
* Request ID
* Tool name and parameters
* Source IP
* Outcome (including whether access was denied)
* The AI client's name and version, reported when it connects

To get the logs, contact Drivetrain Support. Direct export to a SIEM is on the roadmap.

</details>

<details>

<summary>How long do MCP sign-ins stay valid?</summary>

If your company signs in to Drivetrain with SSO, access tokens are valid for 24 hours and your AI client refreshes them automatically for up to 60 days. After that, you sign in again.

With OAuth sign-in, with or without SSO, signing out of Drivetrain or being deactivated ends the connection on the next tool call. A deactivated user's API key also stops working.

If you connect with an API key instead, the key doesn't expire. It stays valid until you select **Reset Key** in **Settings → API Key**.

</details>

<details>

<summary>How do I stop someone from using MCP?</summary>

Make sure none of their roles grants MCP (set **MCP** to **No access** on each, or move them to roles without it), or deactivate the user. This takes effect on their next tool call. Admins always have MCP, so to stop an admin, deactivate them, or have them select **Reset Key** and then remove their admin role, making sure none of their other roles grants MCP. API keys belong to admins: to cut off a key, its owner selects **Reset Key** in **Settings → API Key**, and the old key stops working immediately. See [Set up MCP access](https://integrations.drivetrain.ai/extending-drivetrain/using-the-drivetrain-mcp-server/set-up-mcp-access.md#revoke-access).

</details>

<details>

<summary>Can MCP change my model or my numbers?</summary>

Not on its own. The write tools only create board drafts and dataset transformation drafts. Nothing changes until you review and save them in Drivetrain; a saved transformation then changes the data that feeds your model.

</details>


---

# 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://integrations.drivetrain.ai/extending-drivetrain/using-the-drivetrain-mcp-server.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.
