Agents

Scipio 4.0 exposes every application webapp to an AI agent through the Model Context Protocol (MCP), over Streamable HTTP. Each application has its own endpoint, for example /ordermgr/mcp for orders or /accounting/mcp for accounting; /admin/mcp is the hub that reaches every application from one connection.

The token is a user login#

An MCP access token is not a separate credential type. Every token points at one userLoginId. An agent that connects with a token acts as that user, with exactly that user’s permissions. No permission comes from the token itself. Scipio ships a seed user, scp-agent, in the read-only SCIPIO_AGENT security group, for a first connection.

Default deny#

A call passes only when every gate in the policy engine allows it: server access (MCP_ACCESS, the webapp’s base _VIEW permission), tool access, a service deny list, gateway permission for a direct service call, read-only classification for a read-only token, the service’s own declared permissions, and a component-level base permission for any service the earlier gates did not already decide. The first failing gate denies the call.

One audit row per call#

Every tool call writes one McpAuditLog row, whether it was allowed or denied. Webtools > Agent Access > Audit lists them, with a DENIED filter for blocked attempts.

Read next#

• Quickstart : create a token and connect a client in five steps.
• Security : the permission model, the policy engine, tokens, and the operations checklist.
• Extending agent access : add a tool, a server profile, a provider, or a skill.