feat(agents): Agent chat via MCP — send messages and receive AI responses

[johnxie]

Summary

Add MCP tools for interacting with Taskade AI agents in real-time — send messages and receive agent responses directly from Claude, Cursor, or any MCP client.

Proposed Tools

• agentChatSend — Send a message to an AI agent and receive a response
• agentChatHistory — Retrieve chat history for an agent conversation

Use Cases

• Chat with a Taskade AI agent from Claude Desktop
• Build multi-agent workflows where one AI orchestrates Taskade agents
• Use Cursor to query a Taskade knowledge-base agent while coding

Context

The MCP server currently supports agent CRUD (create, update, delete, publish) and knowledge management. Adding chat capabilities completes the agent lifecycle — create, configure, and interact — all from any MCP client.

Related: Taskade AI Agents documentation at https://www.taskade.com/agents

[johnxie]

Shipped in v0.1.0 (PR #55, 2026-06-27) as part of the API v2 tool layer:

• agentChatSend → promptAgent — send a message to an agent and get its reply (sync request/response, no streaming — matching the v2 API)
• agentChatHistory → listConversations + getConversation

Available on npm as @taskade/mcp-server ≥ 0.1.0 — see the README's Agent Chat & Webhooks (API v2, beta) section. Closing as completed.