MCP Servers
The MCP Servers page manages remote MCP integrations.
It is also one of the clearest examples of the platform/runtime boundary. The platform owns server registration, OAuth posture, and role policy. The runtime later consumes the granted MCP surface when a task is claimed.
What Lives Here
Section titled “What Lives Here”- remote MCP server registrations
- connectivity verification and re-verification
- discovered tool, resource, and prompt snapshots
- shared OAuth client profiles and reconnect flows
- the MCP surfaces operators can later link to specialists
Why It Matters
Section titled “Why It Matters”This page lets operators treat external tool surfaces as explicit policy objects instead of as hidden prompt assumptions.
Registering a remote MCP server does not automatically expose it to every agent. The normal flow is:
- register and verify the server here
- link or grant it to the right specialist on the Specialists, Skills, And Models page
- let the runtime expose that server’s tools only when that specialist is actually executing a claimed task
That split matters. It keeps external tool access deliberate instead of turning every verified integration into a globally available prompt assumption.
Operator Use
Section titled “Operator Use”Use this page when you need to:
- register a new remote MCP endpoint
- repair broken auth or connectivity
- review the discovered capability surface before linking it to specialists
Relationship To Specialists
Section titled “Relationship To Specialists”The Specialists, Skills, And Models page is where remote MCP servers become part of a specialist’s usable tool surface.
Think of this page as the inventory and trust boundary, and the specialist page as the assignment boundary:
- here, you define what external MCP surfaces exist and whether they are healthy
- there, you decide which specialists may use which MCP surfaces
- during execution, the runtime loads only the MCP servers granted to the specialist that claimed the task