Your supervisor fields the same three calls every morning: log this, is the part in, who has that job. Meanwhile the operations team wants its AI assistant to answer the very same questions. A CMMS MCP server for AI agents and an ordinary phone line for people can be two doors into one system instead of two separate projects. OXMAINT AI puts a maintenance request line for people and AI agent CMMS access for Grok and ChatGPT on the same tools, the same role rules and the same activity log, so nothing forks by interface. Book a demo to map your roles to both doors, or start a free OXMAINT AI trial and explore the shared tool set.
CMMS MCP Server or Phone Line: One Maintenance System, Two Doors
- 1Request or inspection
- 2Issue or defect
- 3Work order
- 4Planned maintenance
CMMS MCP Server and Phone Line: Where Front-Door Effort Goes
The interface changes, but the maintenance logic should not. When phone intake and AI agents run on separate tools, someone ends up re-keying the same job. The two illustrative setups below show where that effort goes.
Illustrative: Where Front-Door Effort Goes
Data shown: One system, two doors: Phone requests 35%; Agent requests 20%; Screen work 45%. Separate phone and agent tools: Re-keying between tools 20%; Phone requests 30%; Agent requests 15%; Screen work 35%.
Role Access Matrix for Phone Line and MCP Agents
Use the matrix to see what each role can reach on the phone line, through Grok MCP, or through ChatGPT Actions. The roles stay narrow on purpose, so the same request can move from intake to work without widening the door. Start a free OXMAINT AI trial and test the role matrix
| Action | Technician | Supervisor | Store |
|---|---|---|---|
| Log a maintenance request | Yes | No | No |
| Check stock | Yes | Yes | Yes |
| Read PR/PO/GR status | Yes | Yes | Yes |
| Assign or reassign a work order | No | Yes | No |
| Close technical review | No | Yes | No |
| Post goods receipt or issue | No | No | Yes |
How the Two Door Pattern Fits the Build
The Model Context Protocol project describes MCP as an open-source standard for connecting AI applications to external systems: "Think of MCP like a USB-C port for AI applications." OXMAINT AI uses it for the agent door, while the phone line serves people who never open an AI app. Both reach the same tools through the same server-side checks.
Reference Path for One System, Two Doors
-
Step 1Open the doorA technician dials the line or an agent connects with a secret Bearer token.
-
Step 2Set the roleOn the phone, a shared role code opens the Technician, Supervisor or Store view for that call. On ChatGPT Actions, the code rides on every call.
-
Step 3Call toolsEach door reaches the same tools, and anything outside the role is refused.
-
NowEvery write loggedSuccessful write actions from either door land in the same voice activity log, with the channel and role that made them. Reads and a few write tools are not logged.
Two Doors, One Rulebook: What Holds Up
- 1The two-door model holds up. It keeps the process simple for crews and still gives AI agents a controlled path into the same maintenance actions.
- 2Verify role resets in demo. Check that the role code resets at each phone session and that ChatGPT Actions require a role code on every call.
- 3Watch the refusal behavior. A request outside the role should get a clear refusal, not a guessed answer or a wider tool set.
- 4One caution matters most. In a peer-reviewed 2024 study (Rosilius et al., Multimodal Technologies and Interaction), speech-recognition word error rose from about 3% in quiet to about 37% at 87 dB(A). Follow short confirmed commands, a close-talk headset, and a screen fallback.
Need one workflow for people and agents?
Which Door Should Open First?
Start with the entry point your crew will actually use, then add the other door behind the same role rules. Book a demo to map your door choice to the role views.
-
YesDo those roles need actions beyond logging and checking?
- Yes: expose those roles by phone only if you accept that the code scopes roles, not people.
- No: start with the phone line for requests, stock and status checks.
-
NoDo supervisors already work in Grok or ChatGPT?
- Yes: connect the agent door with a secret Bearer token and a role on every call.
- No: pilot the phone door first, and add the agent door once an AI app is cleared for use.
What the Door Pattern Lets the Team See
The main value is not more reach; it is the same tool path with fewer dead ends. You can compare intake, task movement, and role coverage without changing the underlying work rules.
Illustrative: Door Use During a Pilot, by Week
Data shown (indexed): Phone door, W1 to W8: 85, 85, 85, 80, 80, 80, 80, 80. Agent door, W1 to W8: 70, 70, 75, 75, 80, 85, 85, 90.
Demo checklist
- Confirm the same role code behavior on every door.
- Test refusals when a role reaches outside its view.
- Check that ChatGPT Actions reject calls without a role code.
- Verify the secret Bearer token is treated as a credential.
- Walk one request from intake to the same work path.
How OXMAINT AI Works Alongside Your Existing Systems
OXMAINT AI is designed to sit alongside SAP, Maximo or your current CMMS/EAM, not to replace them. The build uses ERP-style parts records and shared role rules so the maintenance front door stays consistent across people and agents. Book a demo to see the role views next to your stack
- 1
Connect
Use the phone line, Grok MCP, or ChatGPT Actions with the same tool set. - 2
Scope
A shared role code or token scopes the session to Technician, Supervisor, or Store. - 3
Route
Requests, stock checks, assignments, and parts records follow one maintenance path. - 4
Review
Humans decide closes and RCM approvals; the activity log records successful writes.
