Platform · Module
MCP Hub
Expose your One3 Actions to external AI assistants and consume approved MCP tools
MCP Hub works in two directions over the Model Context Protocol. It exposes a tenant's entitled Actions as MCP tools for external assistants with scoped grants, and lets One3 agents consume external MCP servers after review.
Expose: your Actions as MCP tools
MCP tool definitions are generated from the Action registry, including each Action's AI description, examples and schemas (tool generation starts in Wave 0). From Wave 1, each tenant gets an MCP endpoint at mcp.one3.ai/{tenant-slug}. The tool list is filtered by the tenant's entitlements and the grant's scopes.
Access uses MCP client grants: OAuth authorisation code with PKCE per client, scopes chosen by the user (read-only by default), expiry and revoke. Every call executes through the Action Bus with the actor recorded as the user via that MCP client. R3 and R4 Actions create approvals inside One3 and are never approved inside the external client. Users and admins manage grants on one screen showing client, scopes, last use and a revoke button.
Consume: external MCP servers
Planned for Wave 2: an external MCP server registry where admins register a server, fetch its tool list, work through a security review checklist, approve it for the tenant and choose an allow-list of tools. External tool outputs are treated as untrusted and go through the quarantined reader, so they cannot trigger R3 or R4 Actions directly.
Governance
Wave 2 also adds admin policy to allow or deny MCP access per tenant, role and client, with maximum scope and allowed tools; rate limits per grant and plan; and tool usage analytics and audit per client and grant.
Later
Planned for Wave 3: read-only resource and prompt exposure (spec support to verify), discoverability metadata for external agent registries, a platform-curated MCP server catalogue with one shared security review and one-click tenant enablement, tool retrieval so large catalogues stay out of model context, and long-running tool calls. Agent-to-agent protocol support is planned for Waves 5–6. MCP specification status is tracked in the Fact Register.
Features
What MCP Hub includes
Tools from the Action registry
AI descriptions, examples and schemas become MCP tools. W0.
Tenant MCP endpoint
Tool list filtered by entitlement and grant scopes. W1.
Scoped client grants
OAuth with PKCE, user-selected scopes, read-only default, expiry and revoke. W1.
Execution through the Action Bus
R3/R4 create in-One3 approvals; never approved in the external client. W1.
Grant management
Clients, scopes, last used and revoke for users and admins. W1.
Admin MCP policy
Allow or deny per tenant, role and client; maximum scope; allowed tools. W2.
Rate limits
Per grant and per tenant plan. W2.
External MCP server registry
Register, fetch tools, security checklist, tenant approval and tool allow-list. W2.
Untrusted tool outputs
External outputs go through the quarantined reader and cannot trigger R3/R4. W2.
Usage analytics and audit
Per client and grant. W2.
Curated catalogue and tool retrieval
Platform-reviewed servers with one-click enablement; intent search over enabled Actions. W3.
Workflow
Connect an external assistant
- 1
User adds the tenant MCP endpoint to their assistant.
- 2
OAuth with PKCE asks the user to choose scopes; read-only is the default.
- 3
The assistant lists only entitled, in-scope tools.
- 4
A write that is R3 creates an approval in One3.
- 5
The user revokes the grant later from the grants screen.
AI
AI skills and Digital Employees
External MCP server review assistant
Summarises tool-list risk and requested data and pre-fills the checklist; an admin approves. W2.
AI-readable Action descriptions
Help external assistants call One3 Actions correctly.
Digital Employees that use this module
- AI Agent Governance Auditor
They act through the same governed Actions as people; risky steps wait for approval.
Frequently asked questions
Can an external assistant approve risky Actions?
No. R3 and R4 Actions create approvals inside One3, which a permitted human decides.
What does an assistant see by default?
Only read-only tools, filtered by your entitlements and the scopes you grant.
Can One3 agents use third-party MCP servers?
Yes, from Wave 2, after an admin registers and reviews the server and allow-lists its tools.
See MCP Hub in your workflow
Tell us about your team and the outcomes you want. We will walk you through the apps, the Digital Employees and the plan for your industry.