Capabilities & Tool Configuration

Agents are functionally expanded through the assignment of MCP Servers, specific tools, and Skills.

1. MCP Server Assignment

An Agent can be assigned multiple running MCP Servers.

• Effect: The Agent “sees” all tools offered by the server in its system context.
• Update: New functions of a running server can be immediately incorporated into the configuration via the “Update Tool List” link.

2. Selective Tools

Instead of enabling an entire server, you can specifically select individual functions in the “Tools” field. This increases security and reduces token load (shorter system prompt).

3. Tool Approval (Default)

This mode determines how the system reacts when the AI wants to execute an action:

• Request Approval (Default): The user receives a card in the chat and must manually confirm each tool call.
• Full Access: The AI is allowed to execute actions without confirmation (recommended only for trustworthy internal tools).
• Blocked: Tool calls are fundamentally rejected.

4. Bulk Actions

To speed up configuration with many tools, buttons such as “Select All” or “Server-related Selection” are available.

5. Skills

Skills are reusable capability modules that extend the agent with specialized knowledge and workflows. Assigned skills appear in the agent’s system context as a catalog; the agent activates relevant skills via the activate_skill tool.

• Assignment: Select skills in the Skills multi-select field in the agent accordion.
• Scope: Global skills (installed by admin) and user-specific skills are available.
• Effect: The agent sees skill names and descriptions in every run and can load their full instructions on demand.

See Skills — Concept for details.