MCP: examples and workflows

Use these prompts with a connected ChatGPT, Claude or another MCP client. Read Overview and access first. Replace bracketed placeholders with your intended targets; do not paste passwords or tokens into a conversation.

Describe what you want to achieve. For diagnostics, include the symptoms and when the problem started. You do not need to specify tool calls or technical steps. Say if you want the assistant to only investigate or ask before making changes. Prompts are instructions to the model, not a security boundary: grant only the access needed for the task.

For custom script-based checks, see Migrate or create checks with MCP for prompts and the workflow from source to tested check.

Start with a read-only device lookup

The client needs View octoja data. Results stay within your account's device and customer access.

Use the octoja MCP to check whether [device name] is online right now.

Check that the assistant actually calls list_devices. A successful call with no matches is not a connection failure; check the search term and your account's access.

Retrieve patch policies and results

These read-only tools require View octoja data for the client and View Patch Policies and Cycles for your account. Results remain limited to your accessible customers and devices.

Both lists return at most 100 entries by default, or at most 500 with limit. Ask the assistant to identify the policy and cycle first rather than guessing IDs.

A device status of Completed does not mean every update succeeded. Successful update outcomes are Succeeded or AlreadyInstalled; Failed indicates a failure. Outcome counts include only devices that reported results and are visible to you. The detail response lists at most 100 devices with Missed or Incomplete and the ten most frequent errors; status counts cover all visible reported results. A short or empty detail list therefore does not establish success across the entire fleet.

Update device and customer data

Use the octoja MCP to change the alias of [device name] to [new alias].

update_device updates device fields and requires View octoja data, Make changes to octoja and the Edit Device device action. update_customer requires write access and Customer Management. Customer and device access boundaries still apply. With list_custom_fields, the assistant can first discover available custom fields and their types.

Before changing anything, check the target and intended fields. The update tools for devices, customers, cases and custom checks use update objects: only fields with hasValue: true change; value: null clears a field only where it is nullable. Omitted fields stay unchanged. Customer tags replace the entire tag list rather than just adding entries. Ask the assistant to read back the result afterward.

Run or schedule device actions

Use the octoja MCP to schedule a restart of [device name] for tomorrow at 10 pm German time.

A trusted client can restart or shut down devices, or run an existing automation, immediately or at a future time. This requires Make changes to octoja and Run scripts and automations. Your account also needs View Automations and script-execution access on the target devices. Reading definitions and run history additionally requires View octoja data.

1. Ask the assistant to identify the devices with list_devices. For an automation, inspect it with list_automations and get_automation before scheduling.
2. Confirm the target devices and action. Use schedule_device_action with at most 500 device IDs and exactly one of automationId or powerAction (Restart or Shutdown). Supply scheduledFor as a future timestamp with its UTC offset; omitting it runs the action now.
3. Inspect future runs with list_automation_runs and futureScheduledOnly: true. Each device has its own run. Open a run with get_automation_run to check its status and step results; scheduling is not proof of successful execution.
4. To stop an unwanted run, use cancel_automation_run with its run ID. Cancellation does not undo work already completed.

Include the timezone so the assistant schedules the right time. You can also ask it to run an existing automation:

Use the octoja MCP to run [automation name] on [device name]. Let me know whether it succeeded.

Control remote computers

Use the octoja MCP to start a remote desktop session on [device name]. Describe what is happening on screen.

Use the octoja MCP to start a remote desktop session on [device name]. Open settings and switch the system to dark mode.

Use the octoja MCP to start a remote desktop session on [device name]. Open Outlook and delete all emails in the junk email folder. Do not empty the deleted items folder.

The additional Control remote computers authorization lets a trusted assistant view screenshots and operate the mouse and keyboard on an online Windows or macOS device. It is off by default and independent of live data, write access and script execution. Confirm the extra warning only if the client should receive this interactive control. Reauthorize an existing connection if it does not have this grant.

Your Remote Desktop device action and the user-consent and lock-on-disconnect settings still apply. The client cannot replace required consent with its own bypass parameter. On macOS, you need macOS 14 or newer and a user signed in to the graphical desktop. The agent also needs system permissions for screen recording and accessibility; grant them locally through the agent's permissions dialog. See also the Remote Desktop prerequisites.

You do not need to ask for a list of login sessions. The assistant can ask if the target is ambiguous. Its clicks and inputs act directly in the signed-in user's applications. Give it a specific task: the Outlook example deletes messages in the junk email folder rather than deciding which messages in the inbox are spam.

Ask the assistant to end the remote desktop session when finished. Inactive sessions close after about ten minutes. Disconnecting the MCP client also closes its computer-use sessions.

Retrieve diagnostic data reliably

Use the octoja MCP to diagnose performance problems on [device name]. Every afternoon, RAM and CPU usage shoot up. The computer gets slow and choppy, and applications crash. Find out what is causing this. You can read Windows Event Viewer, access log files and run diagnostic scripts, but do not make any changes to the system.

Describe the problem rather than choosing event channels or log files in advance. The assistant can find the relevant sources itself. Live tools require their corresponding device action as well as MCP authorization; seeing a device alone does not grant file access or script execution.

• When reading Windows Event Viewer, the device may reject access to individual channels. The assistant should identify these gaps. A query returns at most 100 events, not a complete event history.
• A file archive returns a download link as both text and a resource link for clients that display it. The link is single-use and valid for five minutes. Download promptly and treat it as confidential; request a new archive after use or expiry.
• For script errors, inspect the execution status and error output. Text output alone does not confirm success. Ask the assistant to report failures rather than interpreting empty or incomplete data as a healthy state.

Create and test Simple SNMP checks

Use the octoja MCP to create an SNMP check for the temperature of [network device]. The OID is [OID] and returns the temperature in degrees Celsius. Warn at 60 °C and mark it critical at 75 °C. Test it on the device before saving.

Ask the assistant to read describe_custom_check_authoring first. With create_custom_check, mode: "Snmp" and snmpConfig, it creates a Simple SNMP check; scripts, input definitions and output fields are generated automatically. The client needs View octoja data and Make changes to octoja, and your account needs Custom Check Management. Prefer the network device's SNMP connection settings over embedding credentials in a shared check definition.

Before saving, test with live_test_script, the target device ID, the executing agent's platform and snmpConfig; omit both script and inputDefinition. The test saves no check. A network device uses its assigned monitoring agent, which must be online. The client needs Run scripts and automations, and your account needs Monitoring Check Management and script-execution access on both target and monitoring device. verified: true confirms valid execution and a parseable result, not necessarily OK status: Warning or Critical can also be valid results.

Inspect the saved check with get_custom_check. A supplied snmpConfig in update_custom_check replaces the entire SNMP configuration; do not send generated scripts or field definitions alongside it.