RAIC MCP Server

What MCP is

Model Context Protocol is an open standard for connecting AI assistants to external systems. Instead of pasting exports into a chat window, the assistant calls a defined set of tools against a server that enforces authentication and authorization. RAIC exposes such a server so assistants can read governance data without any human copying it out of the platform.

Authorization flow

• The MCP client registers and initiates an OAuth 2.1 authorization request against RAIC.
• You sign in and are shown a consent screen naming the client and the scopes it is requesting.
• On approval, the client receives a short-lived access token bound to your user and organization.
• Every tool call is executed as you, in your organization, subject to the same row-level security as the web interface.
• Tokens refresh on the standard OAuth cycle and can be revoked at any time; revocation takes effect immediately.

Tool surface

The exposed tools are a fixed, read-only allowlist of aggregation and lookup operations. There is no arbitrary query capability: an assistant cannot construct SQL, reach a table that is not behind a tool, or write data. Tools cover AI systems and use cases, controls and their effectiveness, the risk register, governance policies, supplier and third-party records, framework coverage, and maturity scores.

Scopes and least privilege

• Scopes are requested at authorization time and shown on the consent screen before you approve.
• A scope can only narrow what your own role already permits — MCP never elevates access.
• Org boundaries are enforced server-side; a token issued for one organization cannot read another.
• Declining a scope simply removes the corresponding tools from the assistant's view.

Audit coverage

Every MCP authorization, token issuance, revocation, and tool invocation writes an audit entry scoped to your organization and mirrored to the platform audit log. Entries record the client, the user, the tool called, and the outcome. Failed authorization attempts are logged too, with the token identifier hashed rather than stored, so you can spot a misconfigured or stale client without exposing credentials.

Administration

• The MCP Admin Console lists registered clients, active grants, and recent tool activity.
• Revoke a client's grant to cut off access without disturbing other integrations.
• Per-client activity view helps you attribute unexpected query volume.
• Access to the console is restricted to administrators.

Good practice

Treat an MCP grant like any other credential: authorize only clients you control, review grants periodically, and revoke ones you no longer use. Because the assistant sees real governance data, apply the same care to where its conversation history is stored as you would to an exported report.