Skip to main content
ComponentsEngine
Availability2026.3 or later
This page is for GermainUX administrators. Once the MCP Server is on, each user connects their own AI client. See Connect an AI Client.
The MCP Server is off by default.

How AI clients sign in

AI clients can sign in to the MCP Server in two ways. You can enable both, and the server tells them apart automatically.

Turn on the MCP Server

1

Open the MCP settings

Open the configuration console and go to Services → REST → MCP.
2

Enable it

Click Enable.
The germain.apm.services.rest.mcp configuration page with the Enable button highlighted

The MCP configuration with the Enable button highlighted

This takes effect immediately on every node, without a restart. Turning it off again takes the MCP endpoint offline straight away. The endpoint URL is your GermainUX address followed by /mcp, for example https://example.cloud.germainux.com/mcp. With OAuth, users connect by entering only the endpoint URL in their AI client. The client then opens a browser window where the user logs in to GermainUX and approves access.
1

Open the OAuth server settings

Go to Services → REST → MCP → OAuth Server and set Enabled to on.
2

Set the issuer

Set Issuer (issuer) to the address users reach GermainUX on, for example https://example.cloud.germainux.com. The MCP endpoint is then <Issuer>/mcp.Leave Login Page (loginPage) and Consent Page (consentPage) at their defaults unless your workspace is served from a different address.
OAuth server configuration with the Issuer field filled in

The OAuth server configuration with the Issuer set

3

Check which AI clients are allowed

Claude, ChatGPT and clients that run on the user’s machine are allowed by default. See Allowed AI clients.
4

Restart the REST service

OAuth settings are only read at startup.
If the OAuth settings are invalid (for example, a missing or malformed Issuer), the MCP Server stays offline. The rest of GermainUX starts normally. The service log explains what to fix.

Allowed AI clients

GermainUX only issues tokens to clients you have allowed. The defaults cover Claude, ChatGPT and clients that run on the user’s machine, so most installations don’t need to change anything here. To allow another hosted AI client, add its client metadata URL or callback URL to the matching list. To block one, remove its entries. For example, remove the chatgpt.com entries to stop ChatGPT from connecting. Turn off Allow Loopback Redirects to block clients that run on users’ machines.

Session lifetimes

Enter these values in seconds.

Use API keys (alternative)

Use a GermainUX API key if your AI client can’t use OAuth, or if you’re connecting an automated agent.
1

Enable bearer authentication

Make sure bearer (API key) authentication is enabled in your authentication settings.
2

Create an API key

Create an API key for the user the AI client should act as from the GermainUX workspace.
The API Key dialog with Name, Expires and permission options

Creating an API key

3

Hand it to the user

Give the user the endpoint URL and the key. They add it to their client as an Authorization: Bearer <key> header.
API keys don’t expire on the short OAuth schedule. Store them like passwords and rotate them regularly.

What users can see and do

  • Every tool call runs as the signed-in user (or the API key’s user). That user’s permissions, team-based visibility and data-masking settings apply exactly as they do in the workspace.
  • Almost all tools only read data. The exception is update_ux_profile_customizations, which changes UX monitoring profile scripts. It requires the configuration-write permission, and its changes only apply to new sessions.
  • search_content is only offered when a content indexer is configured.

Limit which tools are offered

To hide tools from all AI clients, add their names to Disabled Tools under Services → REST → MCP. For example, add update_ux_profile_customizations to make the server strictly read-only. This takes effect immediately, without a restart. See Available Tools for all tool names.

Query limits

These limits protect GermainUX’s query capacity for dashboards and alerting. They apply per server node and take effect immediately. When a limit is hit, the AI client gets a message explaining what to change or how long to wait. It can usually recover by itself.

Monitor MCP usage

Every MCP tool call is recorded in the Audit Log. This includes calls that failed or were stopped by a query limit.
1

Open the Audit Log

In the workspace, go to System → Audit Log and select the MCP Tool Call tab. You need the configuration-write permission to see this tab.
2

Find the calls you're interested in

The list shows each call’s time, tool and user, for the selected time range. Use the search bar to filter by tool or user.
Audit Log screen with the MCP Tool Call tab selected, listing tool calls by time, tool and user

MCP tool calls in the Audit Log

3

Open a call for details

Click a call to see:
  • whether it succeeded or failed, and how long it took
  • the AI client that made the call (for OAuth sign-in)
  • the error, if it failed
  • the arguments the AI client sent. For run_scripted_query, the script is shown separately as code.
  • the result. Small results are shown in full. Larger ones are summarized, for example as a row count rather than the rows themselves.
Audit Log details panel for an MCP tool call, showing status, duration, client, arguments and result

Details of one MCP tool call

Each call is also recorded as an Audit:MCP Tool Call fact, so you can analyze MCP usage like any other GermainUX data.

Next steps

Connect an AI Client

Add GermainUX to Claude, ChatGPT, Cursor, VS Code and more.

Example Prompts

See what you can ask, and how the tools answer.

Get Help

Contact GermainUX Support

The GermainUX team helps you get up and running faster, empower your team, turn data into actionable insights, and optimize performance to deliver measurable business outcomes.

Can't Find What You Need?

Search all GermainUX docs with Ctrl + K, or use the feedback buttons at the bottom of this page to tell us what is missing.

The content of this document is proprietary to GERMAIN SOFTWARE.