AI Agent

On the AI Agent screens, you can view, add, and update AI agents that run on Kuroco. For each agent, you configure the "permitted actions" — the tools, skills, and MCP servers it is allowed to use.

Beta

This feature is currently in beta. Usage is free of charge during the beta period. Each agent session runs in an isolated environment — your data is not used for training and is not exposed externally. However, the agent execution location is in the US. Data resides in the Tokyo region. During the beta period, Diverta Inc. system administrators may review session content for service and performance improvements.

Agent List

How to access

Click [AI] -> [AI Agent].

Field descriptions

Field	Description
Enabled	Indicates whether the agent is enabled.
Name	Displays the agent name. Click it to open the agent editor.
Model	Displays the model used by the agent.
Sessions	Click to open the agent's session list.
Updated on	Displays the date and time the agent was last updated.

Click [Add] to create a new agent.

Agent Editor

How to access

Click [AI] -> [AI Agent], then click the name of the target agent in the list.

Basic information

Field	Description
Name	Set the display name for the agent. (Required)
Model	Select the model to use. (Required) The default is claude-sonnet-4-6. The available models depend on the harness type of the execution environment.
System Prompt	Enter the system prompt that controls the agent's behavior.

Environment and status

Field	Description
Environment	Select the agent execution environment. You can manage network and package settings. Click [Settings] to open the Environment editor.
Status	Disabling makes the agent unavailable.
Autonomous execution	When enabled, the agent can be triggered without human confirmation. If the linked member's (MCP auth member's) email address is empty, it is automatically set to the trigger email address below, and notification emails sent to that address trigger the agent. Tool operations at runtime follow the agent's permission policies (such as the Admin MCP read-only setting).
Trigger email address	Displayed when autonomous execution is enabled. Notification emails sent to this address trigger the agent.
Memo	Enter a memo for management purposes.

Permitted actions

Configure the tools and connections available to the agent, organized in tabs.

A permission policy can be set for tool execution.

Permission policy	Description
Always Allow	Skips confirmation when a tool runs.
Always Ask	Requires approval before each tool runs.

Agent Tools

Select the built-in tools available to the agent. In addition to the [Default Permission Policy], you can set a per-tool permission policy.

Tool	Description
bash	Execute shell commands in the container
read	Read file and directory contents
write	Create or overwrite files
edit	Make targeted edits to existing files
glob	Find files matching glob patterns
grep	Search file contents using regular expressions
web_fetch	Fetch and extract content from URLs
web_search	Search the web for information

Skills

Select the Anthropic-managed skills to enable for this agent.

Skill	Description
xlsx	Excel spreadsheet creation and editing
docx	Word document creation and editing
pptx	PowerPoint presentation creation and editing
pdf	PDF document manipulation and analysis

If Kuroco Skills are provided, they are shown as "Auto-included" and are always attached to the agent.

MCP Server (Kuroco API)

Select a Kuroco API MCP server. The agent can use this API as MCP tools.

Field	Description
MCP Server (Kuroco API)	Select the Kuroco API to connect. Only APIs with dynamic access token (dynamic_token) security are listed. On save, an OAuth IdP and client with target_domain=API are provisioned automatically.
Read-only mode	When enabled, write tools (create, update, delete, etc.) of this MCP server (Kuroco API) become unavailable to the AI agent.
MCP Permission Policy	"Always Allow" skips confirmation for MCP tools. "Always Ask" (default) requires approval before each MCP tool runs.

info

If no MCP-enabled Kuroco API is configured, use [Configure MCP on the API] to set up MCP on an endpoint first. See Model Context Protocol (MCP) Integration with Kuroco for the setup steps.

Admin MCP Server

Exposes Kuroco management controllers, such as topics and member, as MCP tools.

Field	Description
Read-only mode	When enabled, write tools (create, update, delete, etc.) of all selected modules become unavailable to the AI agent.
Modules to Expose	Select the management modules to expose as MCP tools. (Required) The number of available tools is displayed per module.
MCP Permission Policy	"Always Allow" skips confirmation for MCP tools. "Always Ask" (default) requires approval before each MCP tool runs.

Click [MCP tools] to view the tool list per exposed module.

For the Admin MCP server specification, see the MCP Server Reference.

GitHub Repository

Mounts the site's GitHub repository as a resource. To use it, configure the repository integration and PAT in [GitHub settings].

MCP authentication

Field	Description
MCP Auth Member ID	Set the member ID used when accessing the MCP server.
Regenerate MCP Credential	Regenerates the MCP Credential token.

After configuring, click [Update] to save.

Sessions

How to access

Click [Sessions] in the agent list, or [Sessions] at the top right of the agent editor.

Session list field descriptions

Field	Description
Title	Displays the session title.
Anthropic Session ID	Displays the session ID.
Created at	Displays the date and time the session was created.

Click [New Session] to start a new session. Click [Detail] to open the session transcript screen.

Session view

View the session transcript (conversation history) and send messages.

• You can send messages to the agent from the message input field.
• When the agent tries to run a tool whose permission policy is "Always Ask", the session waits for approval (Action Required); choose [Allow] or [Deny].
• Click [Interrupt] to interrupt a running agent.

Notes

• To delete an agent, use [Delete]. Sessions linked to the deleted agent are archived as well.

Related documents

• Environment
• MCP Server Reference
• Model Context Protocol (MCP) Integration with Kuroco
• Integrating with Microsoft Teams - An example of building a Teams bot with an AI agent