KonvoAI Docs
MCP server

MCP tools

Tools exposed by the KonvoAI MCP server for reading and acting on conversations.

All tools run with your KonvoAI identity. Organization-scoped reads and writes enforce the same membership rules as the web app (partners cannot use internal tools).

Read tools

ToolDescription
list_organizationsOrganizations you belong to, with your role.
list_conversationsConversations in an organization; optional status (open, closed, resolved).
get_conversationConversation metadata plus the full ticket event timeline for a streamId.
list_membersMembers of an organization (excluding the partner role).
list_ai_agentsThe organization's AI agents with live status, channels, and schedule.
get_ai_agentOne AI agent's details, including any staged channel/schedule edit.
list_channelsThe organization's channels — pick launch targets for set_agent_channels.
get_agent_replyPoll an agent chat session for the latest reply and its status.

Write tools

ToolDescription
respond_to_conversationSend a text reply (streamId, body; optional clientToken for idempotency).
set_conversation_statusSet status to open, closed, or resolved.
assign_conversationAssign to a user or ai_agent, or pass null to clear.
snooze_conversationSnooze until a Unix ms timestamp, or null to unsnooze.
pin_conversationPin or unpin a conversation.
message_ai_agentSend a message to an agent's configurator chat; returns ids to poll with get_agent_reply.
answer_agent_questionAnswer (or dismiss) a question the agent asked mid-conversation.
set_agent_channelsStage which channels the agent answers on, plus an optional schedule.
publish_ai_agentPublish the agent's draft config — knowledge, tools, staged channels — making it live.
unpublish_ai_agentTake the agent off the air; channels and drafts stay untouched.
create_ai_agentCreate a new (unpublished) AI agent.
rename_ai_agentRename an AI agent.
delete_ai_agentPermanently delete an AI agent and its configuration.

Write tools dispatch through the same server mutators as the KonvoAI web app, including permission checks, billing gates, and outbound send queues for replies. AI-agent management additionally requires a role that can manage operational settings.

Example flow

  1. list_conversations with { "status": "open" }.
  2. get_conversation with { "streamId": "…" }.
  3. respond_to_conversation with { "streamId": "…", "body": "Thanks for reaching out!" }.

Talking to an AI agent

Agent replies are produced asynchronously by the AI worker, so the chat is a send-then-poll flow:

  1. message_ai_agent with { "agentId": "…", "message": "…" } → returns sessionId and replyMessageId.
  2. get_agent_reply with { "sessionId": "…" } — repeat until status is no longer in_progress.
  3. If the reply carries pendingQuestions, respond with answer_agent_question, then poll again.

Changes the agent makes in chat land on its draft configuration; publish_ai_agent makes them live.

Tool results are JSON text in the MCP content payload.

On this page