Purpose
External AI clients can participate in governed enterprise work without becoming a separate automation stack. Enterprise MCP gives approved external AI experiences the same governed Work tools used by the portal and conversation channels.
That means an approved client can begin work, check status, present an Ask or Approval, and retrieve the governed business result. It does not receive a privileged route around Operational Trust.
What this shows: ChatGPT is not the only interface. Copilot, Teams, Slack, and a customer app can reach the same Role Assistant and the same governed result.
What this is not: A separate automation stack per AI product, or a ChatGPT-only path around Operational Trust.
One governed experience
The channel is a place to begin or continue a conversation; it is not the source of authority. Conversation is the daily home. The temporary Review Workspace opens only for rich review. Enterprise MCP extends the same governed experience to approved external AI clients. The request and approval loops: Conversation-first work and Human tasks, approvals, and Why.
Each surface observes the same human interaction contract:
- Start work from a business objective.
- Check the current status and continue only when the work permits it.
- Present an Ask or Approval when a human decision is required — usually in the conversation; rich review uses Work only when needed.
- Record the response through the same human-response path the portal uses (not a parallel channel-only approval scheme) and let the platform evaluate what happens next.
- Use Why? to explain an allow, deny, or wait outcome in business language.
- Read the business result when work completes or stops safely — not a chat transcript as proof.
This consistency lets a person move between a channel, temporary Review Workspace, and an approved external AI client without changing the meaning of a decision. Personal chat memory stays on the external platform; AI Fabrix owns organizational business context only.
What the assistant shows vs full state
External AI and channels receive a presentation envelope: a user-safe view of the current governed work state (waiting for input, awaiting approval, executing, completed, safely stopped, and related states). The envelope helps the client phrase the next useful step.
| Layer | Role |
|---|---|
| Governed work state | Source of truth for status, pending human tasks, and permissions |
| Business result | Validated outcome of the Process Execution when available |
| Presentation envelope | Structured projection for display — does not invent facts or replace the result |
Clients must preserve governed facts, required fields, choices, and approvals — they must not invent enterprise Reality from private chat memory. Architecture: Process Execution Output.
Governed Work tools (product language)
Approved clients use the same categories of tools as Work:
| Capability | What it does |
|---|---|
| Describe / context | Stable method for how the client should talk to AI Fabrix — not a dump of the whole enterprise |
| Start work | Begin governed work from a natural-language business objective or question |
| Status / continue | Read current state and continue when the work permits it |
| Human response | Record an Ask, choice, or Approval for a pending interaction |
| Activity logs | Business-facing activity for the work; optional debug detail only when policy allows |
| Business result | Retrieve the validated result for completed or stopped work |
Treat the tools above as the supported governed set until your deployment adds more projections. Do not assume a private side API.
Human decisions stay governed
External AI can present an Ask, choice, or Approval. It cannot turn a conversation into an ungoverned action. The platform evaluates the response before continuing. Detail: Human tasks, approvals, and Why.
Why Enterprise MCP matters
Organizations often use several AI experiences. One governed surface keeps permissions, work status, human decisions, and Evidence consistent. Enterprise MCP is an integration path, not a replacement for Role Assistants or Work.
Example
A customer service leader asks an approved external AI client to prepare a renewal-risk review. The client starts governed work, reports that an owner’s answer is required, and presents the Ask. If the response requires review across several accounts, the leader opens the Review Workspace, responds in context, and returns to the external AI client. When work completes, the client shows the same business result operators see in Work — not a separate success narrative from chat memory.
Business value
Using one governed tool set across channels, Work, and Enterprise MCP helps organizations adopt approved AI experiences without creating parallel controls or inconsistent human-decision behavior. It preserves the evidence needed to understand what happened, wherever the request began.
For channel delivery, see Assistant channels. For the Work MCP system key and tool catalog, see Runtime Orchestration MCP. For capability keys and Layer A packaging, see Enterprise MCP reference. For business result language, see Business value from work steps.
Limits
Supported external AI clients and their presentation vary by deployment and provider policy. A client may need to open Work for a rich review. Availability of an integration does not remove role, policy, approval, or evidence requirements.