Connections and external tools

Connections let AI Assistant retrieve live information and, when explicitly configured, perform operations in external systems. Open Connections in the RAUM dashboard. The current workspace has three tabs:

• API Connections
• MCP Servers
• Authentication

​

SOURCE versus ACTION

Every tool should be classified according to what it actually does.

​

API Connections

An API Connection can describe a direct HTTP integration, including:

• name and description;
• SOURCE or ACTION category;
• URL;
• HTTP method;
• request headers and query parameters;
• request-body templates;
• variables and input schema;
• authentication;
• output mapping;
• cache duration;
• Simulation behavior; and
• automatic approval behavior for actions where permitted.

​

Authentication methods

The current API Connection editor supports:

• no authentication;
• API key;
• Bearer token;
• Basic authentication; and
• OAuth 2.0 client configuration.

When you have a reusable credential, use the Authentication tab rather than copying the same secret into several connection forms.

​

Shared Authentication

The Authentication tab manages encrypted credentials that can be attached to connections. When a connection uses a selected shared credential, inline secret fields are hidden and the connection references the stored credential instead. This makes rotation and reuse safer than keeping several independent copies of the same secret.

​

Test before saving

The connection editor can run a live test using the same request-building path used by saved connections. Use representative variable values and inspect:

• the destination;
• status/result;
• returned data shape; and
• output mapping.

A successful HTTP response is not enough. Verify that the returned fields actually mean what your procedure expects them to mean.

​

JSON versus web-page sources

API Connections can use structured JSON responses or a web-page retrieval mode where supported. A web-page connection is treated as a read-only source. Use it for fresh information that should be fetched at answer time rather than periodically crawled into the long-lived knowledge base.

​

MCP Servers

RAUM supports hosted remote MCP servers through an HTTP/SSE-style server URL. After connecting an MCP server, RAUM can discover its tools. Tool-level settings can control behavior such as caching, Simulation safety, and approval where supported.

​

Action approval

ACTION execution is fail-safe by default. autoApprove is not enabled by default for connections or discovered MCP tools. When an action is not pre-authorized, the Assistant workflow can stop and present an agent decision rather than silently mutating the external system. Use automatic approval only when the action contract, input validation, idempotency, and business risk actually justify it.

​

Simulation

The safe to simulate configuration is separate from normal action approval. In Assistant Simulation:

• SOURCE tools run normally so the test can retrieve authentic information;
• ACTION tools are intercepted by the dry-run flow and show what would have executed without applying the side effect.

Learn about Simulation →

​

Use connections in procedures

Procedures can reference saved API connections and MCP tools through the procedure editor. That lets RAUM combine a written business process with live evidence or an explicit action rather than burying API behavior in a generic prompt.

​

Organizations using AI Chatbot too

An internal Assistant connection is not automatically public. If your organization also uses AI Chatbot, customer-facing connection access has its own approval and public-access policy. Keep that boundary intact even if the same underlying API is useful to both products.