Audience 01
AI developers
Needs the same permission model for people and agents.
Agents, MCP & capabilities · Concept
Where the work actually changes.
ModelPlugin is aimed at AI developers, model publishers, enterprise AI infrastructure teams.
01
Who it is for
Three audiences, one boundary model.
Audience 02
model publishers
Adopts it to replace a patchwork of tools with one reviewable surface.
Audience 03
enterprise AI infrastructure teams
Adopts it to replace a patchwork of tools with one reviewable surface.
02
Use cases
Each row maps a capability to the operational change it is meant to produce.
| Capability | Operational change | Status |
|---|---|---|
| Model capability registry | Makes an existing action reviewable. | Specified |
| Provider and adapter plugins | Removes an ad hoc step from the workflow. | Specified |
| Inference configuration and routing | Makes an existing action reviewable. | Specified |
| Evaluation and compatibility metadata | Turns a manual control into an explicit contract. | Specified |
| Standardized APIs, SDKs and MCP surfaces | Removes an ad hoc step from the workflow. | Specified |
03
Migration
How a team would move from the status quo without a big-bang cutover.
Step 01
Map the current path
Name the tools, owners and handoffs that the product replaces. Nothing is removed yet.
Step 02
Run one capability
Pick a single scoped capability and run it alongside the old path, with evidence compared side by side.
Step 03
Retire the old path
Remove the previous tool only when the new record is complete enough to audit.