Node.js · Express.js · Model Context Protocol
AI Systems · regular
Unified Agent Context System is a TypeScript/Node MCP server: it registers tools under configurable namespaces, supports streamable HTTP and stdio transports, and serves content (e.g. profile, projects) from files with path-safety. It demonstrates a layered MCP implementation suitable for deployment (e.g. containers).
Build a modular Model Context Protocol (MCP) server with a separated core, transport layer (HTTP / stdio), and configuration-driven tool modules so agents and IDEs can access structured context (e.g. Markdown-backed portfolio) in a safe, predictable way.
The project assumes the MCP standard as a stable contract between agents and server capabilities, instead of ad hoc message formats. The architecture is layered: a shared core (MCP server instance, shared types, tool registry), a pluggable transport (streamable HTTP by default, stdio as an alternative), and domain modules loaded from configuration with isolated tool namespaces. Modules are not hard-coded into the main entrypoint—enablement, options, and content roots come from config and environment variables. Portfolio content is read-only, stored in files (Markdown + manifest), with safe path joining (path traversal protection). Operationally, the system should be deployable as an HTTP service (e.g. in a container) with configurable host/port and an optional internal bearer token, while keeping a clear separation of concerns across layers.
The project gives agents a single, standards-based entry point to curated data and operations instead of scattered scripts or one-off integrations. The design splits server construction, transport choice, and domain logic into swappable modules. The portfolio module illustrates a read-only Markdown corpus with search and section-oriented tools.
Gallery
