Overview

The Platform MCP Server is a service that securely connects a tenant's AI applications with the Intelligent Risk Platform and developer resources.

The Platform MCP Server transforms Platform API specifications, documentation, and code samples into a single, queryable knowledge graph that enables Intelligent Risk Platform tenants to discover and implement catastrophe modeling and reinsurance workflows.

The Platform MCP Server is available to every Intelligent Risk Platform tenant. All that you need to do is assign a principal (API key or user account) to a user group that has been assigned the IRP Agentic Tools User role. The API key or bearer token is specified in the MCP client configuration and passed to the Platform MCP Server to initiate each session.

Security

All tenant data remains on the Intelligent Risk Platform. Data is not duplicated, copied, or moved to an AI environment outside the Intelligent Risk Platform.

Tenants can control access to the Platform MCP Server and restrict the actions that may be performed by AI agents by means of security credentials. Security credentials enable the Intelligent Risk Platform to authenticate the identity of the client application and confirm that the client application is authorized to access and leverage the requested resources.

The MCP client must pass valid security credentials whenever it connects to the Platform MCP Server or performs actions using Platform API operations. These credentials are passed separately. The client must pass connection credentials to connect to the Platform MCP Server and execution credentials to use Platform API operations.

Connection Credentials

Connection credentials are security credentials that enable the client to connect to the Platform MCP Server. These credentials are specified in the MCP client configuration and passed to the Platform MCP Server whenever it initiates a new session.

Connection credentials enable the client to use all discovery tools (e.g. search-api-spec, graph-*, get-operation-spec, vector-search, workflows, docs).

Connection credentials do not enable the client to perform actions using Platform API operations.

Execution Credentials

Execution credentials enable the MCP client to use API execution tools to perform read and write actions using Platform API operations. The execution credential must be specified separately from the connection credential. MCP clients cannot read security credentials from the client configuration file.

In chat-based MCP clients, an execution credential must be specified in each session. Execution credentials are managed and tracked as IRP_AUTH session variables. These variables can be set in the chat session using the session-variable tool. Once set, the IRP_AUTH session variable can be passed to use the execute-api tool.

The IRP_AUTH session variable identifies same credential (e.g. API key) that specified in configuration.

Credential Security

Moody's Insurance Solutions recommends that you practice key-safe options for managing and submitting security credentials.

The Platform MCP Server requires that MCP Server clients that intend to use the execute-api tool must submit a valid security credential in every chat session.

Security credentials entered into a chat window become part of that conversation's record. The security credential is stored by the client application, synced with the conversation, and are visible to the AI model. For this reason, many organizations prohibit developers to enter security credentials such as API keys in AI conversations.

Platform MCP Server Role-Based Access Controls

The Platform MCP Server leverages the Intelligent Risk Portal data access and security model.

AI tools have read or write access to tenant resources based on assigned entitlement and role-based access controls. Every request to the Platform MCP Server must pass valid credentials (API key or bearer token) and be authenticated at Moody's API gateway. The API key or bearer token represents a principal that has been assigned the appropriate access rights and privileges.

Every principal that has access to the Platform MCP Server must belong to a user group that has been assigned the IRP Agentic Tools User role.

The IRP Agentic Tools User role grants the principal permission to perform two actions:

The IRP Agentic Tools User role does not grant the principal permission to perform any read or write actions. To perform specific actions, the principal must be assigned the appropriate role. For example, a principal assigned the IRP Agentic Tools User role cannot model exposures unless it is also assigned a role with that permission, e.g. the Cat Modeler role. To learn more about Intelligent Risk Platform™ access controls, see Access Controls.

🍊

Read-Only Roles

The Intelligent Risk Platform does not support a read-only role. All predefined roles support a combination of READ and WRITE actions.