MCP Dust Server

The Ma3U MCP Dust Server bridges the Model Context Protocol (MCP) with Dust.tt’s agentic AI platform, giving AI assistants a direct, low‑latency channel to execute custom agents and retrieve real‑time results. By exposing a full MCP lifecycle—, , , and —the server allows developers to treat Dust agents as first‑class tools within any MCP‑compliant workflow, whether they are building a new chatbot or extending an existing one.

At its core, the server is split into two tightly coupled components. The MCP Server implements JSON‑RPC 2.0 over both SSE and HTTP Stream Transport, handling session state, context preservation, and heartbeat checks to keep connections alive. It also masks PII in logs for security compliance and offers automatic reconnection logic so that transient network hiccups do not interrupt a user’s conversation. The MCP Test Client is a lightweight web interface that mirrors real client behavior, enabling developers to debug streaming responses and verify session handling before integrating the server into production.

Key capabilities include:

• Real‑time streaming of agent responses via Server‑Sent Events, ensuring that users see partial outputs as soon as they are available.
• Direct tool execution through the method, allowing a client to invoke an agent without managing the full message loop.
• Comprehensive health checks and status endpoints, which simplify monitoring in containerized or cloud environments.
• Secure API key handling that prevents accidental exposure of credentials in logs or error reports.

Typical use cases span from enterprise chatbots that pull data from internal systems (e.g., Jira, Salesforce) to customer‑facing assistants that need to run complex multi‑step workflows. For example, a support desk could route a user’s query to a Dust agent that queries Zendesk tickets, updates Slack channels, and returns a concise summary—all within the same MCP conversation. Because the server speaks standard MCP, it can be plugged into any assistant that supports the protocol, such as Claude Desktop or Windrift IDE, without custom adapters.

In short, the Ma3U MCP Dust Server turns Dust.tt’s powerful agentic AI into a plug‑and‑play component for developers, providing robust streaming, secure session management, and seamless integration with existing MCP workflows.