Tool reference

Your client only sees tools for which the token has all listed scopes. Changes affect your account directly; let the assistant show you write and delete actions before they run.

Voice agents

Tool	Scopes	What it does
`list_voice_agents`	`voice_agents:read`	List your agents
`get_voice_agent`	`voice_agents:read`	Read one agent by ID
`deploy_voice_agent`	`voice_agents:write`	Create an agent (name, optional prompt, voice and knowledge-base settings). Does not connect a phone number
`update_voice_agent`	`voice_agents:write`	Change name or description
`delete_voice_agent`	`voice_agents:write`	Delete an agent

Settings

Settings are grouped into prompts, voice_config and knowledge_base.

Tool	Scopes	What it does
`get_voice_agent_concern_schema`	`voice_agents:read`	Read a group’s schema: keys, types, allowed values
`get_voice_agent_concern`	`voice_agents:read`	Read a group’s current values
`update_voice_agent_concern`	`voice_agents:read`, `voice_agents:write`	Change one setting; the response lists side effects
`set_agent_prompt`	`voice_agents:read`, `voice_agents:write`	Change one setting in the `prompts` group

Tools

Tool	Scopes	What it does
`list_tool_types`	`tools:read`	Types that can be created, with hints
`list_tools`	`tools:read`	Your tools; with an agent ID, whether each is active for it
`get_tool`	`tools:read`	Read one tool
`create_tool`	`tools:write`	Create a tool
`update_tool`	`tools:write`	Change a tool’s definition
`enable_tool` / `disable_tool`	`tools:write`	Switch a tool on or off for an agent
`delete_tool`	`tools:write`	Delete a tool; only `external:api` tools
`execute_tool`	`tools:execute`	Run a tool for real, for testing

Conversations

Tool	Scopes	What it does
`list_conversations`	`conversations:read`	Past conversations, filtered by date, number, channel or call result, paginated
`get_conversation`	`conversations:read`	One conversation with its transcript, where retention keeps it

list_conversations accepts page, perPage (1-50), fromDate, toDate, number (agent ID or phone number), callerNumber, searchText, channel (one value or an array), and result. searchText does not search transcripts. The response returns one page with has_more; use fixed date bounds and deduplicate by id across pages. Dates without an offset use Europe/Berlin time. get_conversation takes conversationId, either a numeric ID or the public conversation_id from the list. Transcript omission respects retention rules; no hidden/technical messages or audio recordings are returned. Both tools share the account’s API quota, so avoid continuous polling. Evaluation tools (evals:*, eval_runs:*) also exist; they are not covered in this guide yet.

Prompts

Prompt	Content
`prompting-guide`	The German Prompting Guide
`schema-guide`	The English API Reference → Set up a voice agent

Both also tell the assistant which MCP tools your token allows. See Guides as MCP prompts for how to invoke them.

Getting started

Build

Reference

Voice agents

Settings

Tools

Conversations

Prompts

​Voice agents

​Settings

​Tools

​Conversations

​Prompts

Voice agents

Settings

Tools

Conversations

Prompts