Connect RAIC to ChatGPT, Claude, Cursor, Codex, and Perplexity Comet.
RAIC ships a first-party MCP server so any OAuth-capable AI client can act on the customer's AI governance data as the signed-in user, under their tenant's RLS and role.
Why a first-party MCP server matters
Ask, don't click: "What high-risk AI systems have no owner?" — answered in the assistant, not on a dashboard tour. Analysts and executives get answers where they already work.
Tenant-isolated by design: Every tool call runs under the signed-in user's org via row-level security. The MCP server never bypasses roles, tenants, or approval workflows — it is another authenticated caller, not a backdoor.
DCR-ready: OAuth-capable clients register themselves via Dynamic Client Registration. Admins enable MCP per org from the RAIC Admin Console — no manual client provisioning, no shared secrets to distribute.
Standards over shims: OAuth 2.1 authorization-server metadata and RFC 9728 protected-resource discovery — not copied session tokens, not vendor-only extensions. Any conformant MCP client connects.
Endpoint & discovery
The canonical endpoint and OAuth metadata live in RAIC's discovery documents. Link the discovery URL — don't hard-code the endpoint into your client.
- MCP discovery
- https://app.rhindoncyber.com/.well-known/mcp.json
- OAuth protected-resource metadata (RFC 9728)
- https://app.rhindoncyber.com/.well-known/oauth-protected-resource
- Transport
- Streamable HTTP · protocol version 2025-06-18
- Auth modes
- OAuth 2.1 with Dynamic Client Registration for interactive clients — plus long-lived
rai_*bearer tokens for headless connectors, CI, and back-office scripts.
Canonical MCP server URL, transport, protocol version, and scope catalog. Fetch this first.
Authorization server, supported scopes, and DCR endpoint. Clients use this to bootstrap the flow.
Scope catalog
Scopes map to the same resources you already administer in RAIC. Requests over the MCP server obey the same row-level security and role-based access as the app itself.
| Scope | What it covers |
|---|---|
| openid, email, profile | Identity of the signed-in user. |
| risks:read / risks:write | AI + Cyber risk register with inherent vs residual. |
| controls:read / controls:write | Unified AI + Cyber controls register. |
| systems:read / systems:write | AI + Cyber systems registry. |
| use_cases:read / use_cases:write | AI use case intake and workflow. |
| policies:read / policies:write | Governance policy CRUD (new). |
| audit:read | Tenant-scoped platform audit log. |
What an AI client can do over MCP
Query the register: List AI Systems, Use Cases, Controls, Risks (inherent and residual), Policies, and audit events — filtered by owner, framework, or lifecycle stage.
Take governed actions: Create or update policies, register a new AI system, open a risk, or advance an approval — all under the caller's role and workflow permissions.
Answer executive questions: Assemble narratives from live tenant data: "What changed in our EU AI Act posture this quarter?", "Which use cases have residual risk above tolerance?".
Stay inside guardrails: Assistants can only read and write what the connecting user could already read and write in the app. Approvals, delegations, and audit logging still apply.
Enable MCP for your tenant.
OrgAdmins turn on the MCP server per tenant from the Admin Console. Point your AI client at the discovery URL and complete the OAuth consent screen.
Turn on MCP per org from the RAIC Admin Console.
Discovery, OAuth flow, scopes, and worked examples.
See RAIC and its MCP server in a live tenant.