Why an IT desk wants NinjaOne in ChatGPT
Priya Shah runs the service desk at a managed services provider. Each client is an organization in NinjaOne, and her technicians already write ticket replies in ChatGPT. Their questions are short. Which devices at one client went offline over the weekend? What alerts fired overnight? Which version of an app is on Jake Morgan's laptop? NinjaOne in ChatGPT puts those lookups in the chat where the reply gets written.
MCP (Model Context Protocol) is the standard way an AI assistant calls tools in other apps. NinjaOne documents its own MCP server as an early access feature, with Microsoft Copilot as its worked example (ninjaone.com, checked November 2026). The last section covers when that server is the better fit.
Elaichi's ninjaone connector is the other route. It sits behind the same endpoint as the desk's other apps and can sign in with a NinjaOne client app. The IT lead has two decisions to make: which NinjaOne scopes the client app holds, and which of the connector's writes the desk's role can reach.
Which NinjaOne scopes should the client app hold?
Start with Monitoring, the narrowest scope, and widen only if a test shows you must. A NinjaOne client app offers three scopes, and each one covers all Public API resources of its type (ninjaone.com, checked November 2026). Monitoring grants read-only access to monitoring data and organization structure. Management allows changes to device and organization information, including new organizations, new devices and running scripts. Control turns on remote access through the API.
Only system administrators may configure OAuth tokens for NinjaOne's Public API (ninjaone.com, checked November 2026). The path is Administration, then Apps, then API, then the Client app IDs tab, then Add client app. NinjaOne's own guide to an API server picks API Services (machine-to-machine) there, with the Client credentials grant and no redirect URL (ninjaone.com, checked November 2026). NinjaOne then generates a client ID and a client secret.
NinjaOne's API server guide is plain about the stakes (ninjaone.com, checked November 2026). Depending on the scopes chosen, whoever holds the credentials "can read and write data across your entire NinjaOne environment." For a managed services provider, that environment holds every client organization. NinjaOne's advice is to grant only the scopes the automation requires.
NinjaOne does not document which API reads each scope covers. So tick Monitoring first, then test every read the desk needs. The first test may show that a read needs a wider scope, and that is the moment to decide, read by read. The scope is the outer wall, because it binds every call through the connection, whatever Elaichi's rules would allow.
What the ninjaone connector reads for a device question
Most of what Priya's desk asks is a read. The connector's reads cover devices, alerts, installed software and organizations. list_all_ninjaone_devices returns each device's organization, location, offline status, last contact time and policy. list_all_ninjaone_search_devices finds devices that match a search term. list_all_ninjaone_organizations lists organizations with their locations and policy mappings, which for a provider is the client list.
list_all_ninjaone_alerts lists active alerts with each alert's device, message, source and create time. Device-level reads go deeper. list_all_ninjaone_device_activity_logs shows what happened on one machine, and list_all_ninjaone_device_software shows what is installed. Patch history, disks and alerts for a single device follow the same pattern. Each of those takes one device_id, so the model searches for the device first and then reads its detail.
Fleet-wide reads answer audit-style questions across clients, once a test confirms the client app's scope reaches them. list_all_ninjaone_device_volumes reports storage volumes and BitLocker status. list_all_ninjaone_antivirus_status, list_all_ninjaone_device_health and list_all_ninjaone_operating_systems cover the rest. list_all_ninjaone_tickets runs a NinjaOne ticket board and needs a board_id.
Which NinjaOne writes should a lookup desk never reach?
All of them. The connector's writes change device, organization and end-user records, and one tool deletes an end user. A technician who needs those changes can make them in NinjaOne itself.
| Tool | What it changes | For the desk's role |
|---|---|---|
update_a_ninjaone_device_by_id |
Display name, role, policy, organization, location, warranty | Restrict |
create_a_ninjaone_organization, update_a_ninjaone_organization_by_id |
Organizations, including how new devices are approved | Restrict |
create_a_ninjaone_end_user, update_a_ninjaone_end_user_by_id |
End users, including full portal access | Restrict |
delete_a_ninjaone_end_user_by_id |
Removes an end user | Restrict |
| Device, alert, software and organization reads | Nothing | Allow |
The device update deserves the most thought. It can change a device's organizationId and its policyId, and the tool's schema describes the policy as the device's monitoring and management configuration. On a provider's account, one sentence in a chat could move a machine into another client's organization.
The end-user create takes a sendInvitation option, which invites the new end user, and a fullPortalAccess field. Both are access decisions, not lookups.
What the connector leaves out matters too. No ninjaone tool runs a script, reboots a device or starts a remote session.
Block rules on the desk's role
Block rules on the desk's role keep the NinjaOne device update, organization writes and end-user writes and delete out of reach, and leave the role's other apps untouched. A restriction is a rule naming which connectors and which individual tools a target may reach. In Elaichi, restriction targets are role or user only; the organization default is the absence of a rule, which means allow everything. So the desk needs its own role, and each technician holds exactly one.
Name the device update, both organization writes, both end-user writes and the end-user delete. In Elaichi, a layer with only block rules has no allowlist, so everything it does not block passes it. Slack, the ticketing app and every NinjaOne read stay reachable without being named.
An allow rule is the other shape, and it carries a trap. In Elaichi, once a role carries any allow rule, every connector and tool none of its allow rules names is blocked for that role. An allow rule naming only NinjaOne reads would cut the desk off from every other app it uses. Within a layer a block beats an allow, so a later widening of the role leaves the tools under block rules restricted. Blocks match the tool's name or its operation, while allows match only the operation, as the name versus operation rule explains. A restriction change takes effect within about two minutes.
One client app, one shared connection
An IT admin connects the ninjaone connector once with the client app's ID and secret, then shares the connection with the desk at use. The connector accepts OAuth or OAuth client credentials. With client credentials, every call through the connection uses the client app's credentials, whoever asked the question. NinjaOne's API server guide says machine-to-machine authentication separates automation from individual users (ninjaone.com, checked November 2026). Elaichi then records which member made each call.
In Elaichi, a shared connection runs on its owner's credential. The NinjaOne secret is kept in a separate credential service, encrypted at rest. Choose a connection owner who will stay. If that person leaves, offboarding in Elaichi shows the connections they own, so an admin can transfer this one before removal. The wider checklist for a departing admin applies here too.
Keep the two credentials apart. The NinjaOne client ID and secret belong to the connection. Each technician's sign-in from ChatGPT to Elaichi is a separate OAuth grant, which that person approves in a browser. In Elaichi, revoking a share or disconnecting an account takes effect on the caller's next call. Also in Elaichi, removing or suspending a member revokes every live grant in the same transaction as the membership change.
Publishing Elaichi in the IT workspace's ChatGPT
A ChatGPT workspace admin adds Elaichi once, as one custom app, and the desk signs in to it. OpenAI says custom MCP server connections and full MCP support are available for ChatGPT Business and Enterprise/Edu on ChatGPT web (help.openai.com, checked October 2026). On Business, only admins and owners create and publish custom apps for the workspace. On Enterprise and Edu, admins can also give selected members the Create plugins with MCPs permission, but only admins and owners publish.
Admins and owners create the app from Workspace settings, then Apps, then Create (menu path checked October 2026). Enter https://api.elaichi.ai/mcp exactly, click Scan Tools, complete the OAuth prompt and click Create. On Enterprise and Edu, the Publish dialog's Configure Actions step selects the actions the app may use, and Configure Access picks the groups that get it (help.openai.com, checked October 2026). After an admin first approves the app, ChatGPT works from a frozen snapshot of its tools. Later changes wait until an admin clicks Refresh, and new actions arrive disabled by default.
In Elaichi, connected tools are never listed one by one, however few there are. ChatGPT finds NinjaOne tools with search_tools and runs them through Elaichi's own tools, such as execute_tool and read_tool. OpenAI's admin controls let a workspace turn an app on or off and set its actions (help.openai.com, checked October 2026). Those settings act on Elaichi's tools as wholes and cannot single out one NinjaOne tool. The client app's scope and the role's block rules are the controls. Connecting Elaichi to ChatGPT walks through the sign-in.
Checking the desk's access before rollout
Test from a desk account before anyone relies on it. Sign in as a member on the desk's role and ask ChatGPT which devices at one client are offline. The read should run. Then search for the end-user and organization writes. A restricted tool is left out of the tools Elaichi offers a client and cannot be called. Search still names it, flagged restricted, with no schema.
Then read the audit trail. Elaichi writes one entry for each connected-tool call that reaches execution (a call refused earlier writes none). In Elaichi, a call from an MCP client is recorded under the person who signed in, with surface mcp and the OAuth client named. Each entry also holds the one path argument that names the object, as the target id, and nothing else about the arguments. For a per-device read such as list_all_ninjaone_device_software, that path argument is the device id. A review guide for AI audit rows covers what to look for.
A technician who later needs a restricted NinjaOne tool files an access request. Only a person in the Elaichi console can approve it, under Governance, then Access requests. Approval creates an access grant for that technician alone, and the desk's role keeps its rules unchanged.
When NinjaOne's own AI is the better fit
If the desk works inside NinjaOne all day, NinjaOne's own assistant may cover it. Katana Agentic AI is an early access assistant inside the NinjaOne console. Its listed features include alert and activity summarization, ticket summaries, device health analysis and script recommendations (ninjaone.com, checked November 2026). System administrators have it by default, and access is granted to individual technicians or to technician roles.
For an outside AI client, NinjaOne's MCP server sits at https://{instance}/api/v2/mcp/, in early access. Its article says "we only support the monitoring scope for the MCP" (ninjaone.com, checked November 2026). Admins turn it on under Administration, then Apps, then NinjaOne MCP, and choose which NinjaOne products it may reach. NinjaOne's worked example uses Microsoft Copilot as the host, so test it with the client your team uses. For a team that only needs NinjaOne, that is less machinery.
Elaichi is the better fit when a single question spans several apps. Priya's ticket sits in the help desk, the device in NinjaOne and the escalation thread in Slack. Elaichi gives ChatGPT, Claude, Cursor and Microsoft Copilot the same address, with one rulebook per role and one audit trail across those apps. The same pattern for Apple devices is in Jamf in Claude for a help desk, and for Windows fleets in Intune lookups in Claude. If one person only wants to try a few queries, you may not need a gateway yet. Browse the 900+ connectors or use cases by team.