MCP Server

What profullstack/mcp-server MCP server does

The profullstack/mcp-server MCP server is a Node.js application that exposes standardized HTTP endpoints for server status, health checks, metrics, model discovery, model activation, and inference. It uses a modular architecture so additional functionality can be organized and loaded as separate modules. Module metadata can be returned through the API, and modules can manage their own dependencies through package metadata.

The model layer supports providers and model types described in the project documentation: OpenAI text generation and Whisper speech-to-text, Stability AI image generation, Anthropic text generation, and custom Hugging Face models. The server can perform inference using the active model or a model selected directly in the request.

How it works

The application runs as an ES Modules Node.js service and listens on port 3000 by default. Its standard routes include endpoints for listing models, retrieving model details, activating or deactivating a model, and submitting inference requests. A client can first activate a model with configuration such as temperature, then send a prompt or generation request to the inference route.

Streaming inference is available for compatible models. The repository also includes a module search function, logging utilities, and a test structure covering core behavior and individual modules.

The profullstack/mcp-server MCP server does not provide a global authentication layer. Any client able to reach the listening port can call unprotected routes, so deployments exposed beyond localhost should use an authenticating reverse proxy. Some side-effecting modules accept bearer or API-key tokens, while filesystem and outbound-request modules apply additional restrictions.

Setup and configuration

The project requires Node.js 18 or later and pnpm 10 or later. Installation uses the repository's dependency workflow: install dependencies with pnpm, then start the service with the project's start script. A development script enables automatic reloads. Docker and Docker Compose configuration are also included, with the container exposing port 3000.

Copy the sample environment file to .env before configuring provider access. OpenAI, Stability AI, and Anthropic each use their own API-key variable: OPENAI_API_KEY, STABILITY_API_KEY, and ANTHROPIC_API_KEY. Only configure the provider keys for the models you intend to use.

Optional security-related settings include SCANNER_API_TOKEN, README_BADGES_API_TOKEN, README_BADGES_ROOT, CONVERT2DOC_ALLOWED_HOSTS, CRAIGSLIST_ALLOWED_HOSTS, CSRF_PROTECTION_ENABLED, and CORS_ORIGINS. These control selected module tokens, filesystem boundaries, outbound host allowlists, cross-site request protection, and trusted browser origins.

Tools and capabilities

• Report server health, status, and metrics.
• List available models and inspect model metadata.
• Activate or deactivate the current model.
• Run text, image, or speech-to-text inference where the selected provider supports it.
• Load modules dynamically and manage module-specific dependencies.
• Run the included Mocha and Chai test suites.

Limitations and notes

The profullstack/mcp-server MCP server requires users to supply provider credentials for external model APIs. Those providers may impose their own usage costs, quotas, and capability limits. Streaming is not guaranteed for every model.

The README describes HTTP routes and local deployment, but it does not provide a confirmed one-line package-manager command for installing the project as a standalone MCP client package. The repository must therefore be obtained and installed as a Node.js project. Exposing the service directly to an untrusted network is discouraged because global authentication is not enabled by default.