Model Context Protocol (MCP) Servers
The Model Context Protocol (MCP) is an open standard that enables AI models to securely access external tools and data. MUJPI can act as an MCP Client, routing requests through a Policy and Approval Gateway to approved MCP Servers running locally or remotely. MCP is not a MUJPI product or API vendor; it is an open architecture we support for flexible tool integration.
Architecture & Data Flow
MUJPI communicates with MCP servers via a local or remote transport layer, strictly gated by policy.
Data Flow
Data is passed contextually during inference via the MCP protocol. Local MCP servers retain data on-premise, while remote servers rely on encrypted transport. Audit logs capture tool invocation and responses.
Typical Use Cases
- ✦Querying internal databases via local MCP servers
- ✦Integrating specialized enterprise APIs
- ✦Running secure sandbox environments for code execution tools
- ✦Connecting private file systems to intelligence models
Capabilities
- ✦Standardized tool discovery
- ✦Context and resource fetching
- ✦Prompt template management
- ✦Secure bidirectional communication
Security Considerations
- All MCP tools require explicit policy approval.
- Human-in-the-loop (HITL) gates can be enforced for sensitive MCP tools.
- Local MCP servers ensure data never leaves the edge device.
- Full audit trails of MCP tool execution.
Setup & Deployment
Deployment Modes
Setup Overview
Deploy an MCP server container or script, register its transport endpoint in the MUJPI admin panel, and assign it to an active agent or intelligence route.