> ## Documentation Index
> Fetch the complete documentation index at: https://docs.altrum.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# MCP servers

> Give each MCP server an owner and a status, see what each of its tools can do, and block a server so no key can use its tools.

MCP servers give models tools: search tickets, read a repository, update a record. This page is where your organization decides about them: who answers for each server, whether it is approved, and which of its tools can do damage. Blocking a server switches it off for every key.

Open it from **AI assets**, **MCP servers**. Adding a server and setting how it signs in happen elsewhere, under **AI Gateway**, **MCP servers**: see [MCP tools](/ai-gateway/mcp.md).

| | Owner | Admin | Governance Lead | Auditor |
| - | - | - | - | - |
| See MCP servers | Yes | Yes | Yes | Yes |
| Set a server's owner, description and tool risks | Yes | Yes | No | No |
| Approve, review or block a server | Yes | No | Yes | No |

## Read the list

Each row shows the server and its address, its owner, its status (**Approved**, **Under review**, **Blocked** or **Not reviewed**), how many tools it has, how many are destructive, and whether it is connected. **Needs review** shows the servers nobody has approved or blocked.

## Review a server

Choose a row. Give the server an **Owner** from your members and a **Description** of what it is for, so a reviewer can decide. Then choose **Change status** to approve it, keep it under review or block it, with a note if you want one.

A server added through Altrum starts **Under review** and works straight away. Until it is blocked, its status records a decision and changes nothing about what it does.

## Block a server

Set the status to **Blocked**. Every key stops seeing the server's tools at once, and a call to one of them is refused. It cannot be switched back on under **AI Gateway** until someone approves it or puts it back under review here.

If the AI Gateway cannot be reached, the block is not saved either, so a server never shows **Blocked** while its tools still work: "The AI Gateway could not be updated, so nothing changed. Try again shortly."

## What each tool can do

Every tool has a risk:

| Risk | Meaning |
| - | - |
| **Destructive** | It may delete or change something in a way that cannot be undone. |
| **Write** | It changes things, but not destructively. |
| **Read** | It only reads. |

Altrum takes the risk from what the server says about its tools. A tool the server says nothing about counts as **Destructive** and is marked **Assumed**, because nothing says it is safe. You can set a tool's risk yourself; it then shows **Set by a person** and stays that way when the server lists its tools again. **Use the server's** goes back.

A risk ranks and records; it does not limit which tools a key may use. To limit tools, see [MCP tools](/ai-gateway/mcp.md#choose-who-can-use-which-tools). **View tool calls** opens the server's calls in the [request log](/ai-gateway/request-logs.md), and [Reporting](/reports.md#tool-use) ranks destructive tools first.

Every change is recorded in the [audit trail](/audit-trail.md). Each person can make up to 60 changes to MCP servers an hour.


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.