Tools & MCP — Later capabilities¶
Target
This page describes intended behaviour or structure. No implementation is claimed unless a section is explicitly marked Current implementation or Evidence.
| Capability | Purpose | Trigger | Owner |
|---|---|---|---|
| Cross-organization MCP registry federation and discovery | find and trust servers beyond the tenant | radar item moves from assess to trial | BC-08 |
| Certified MCP servers and tool packs in the Marketplace | distribution of qualified capabilities | Marketplace platform live | BC-27 with BC-08 |
| Computer-use and browser agents at scale | UI-driven tasks as sandboxed tools with recordings | pilot evidence on reliability | BC-08 with BC-26 |
| Tool-call reliability scoring per tool and agent | feed the evaluation gates (D32) and route around flaky tools | evaluation feed in place | BC-22 with BC-08 |
| Per-tenant tool allow-lists and quotas in the Studio | self-service governance of capabilities | multi-tenant pilot | BC-18 with BC-08 |
| Dry-run of whole workflows against descriptors | predict approvals, classes and sandboxes before a run | workflow engine simulation | BC-38 with BC-08 |
| Automatic binding proposals from schema similarity | faster onboarding of consumed servers | registry volume | BC-08 |
| Outbound A2A calls as tools of class external | delegate to partner agents under policy | first partner agent (ADR-0023) | BC-08 with BC-04 |
| Executor profile inside the customer's network for hybrid | sandboxes next to the code | hybrid pilot | BC-08 with BC-25 |