MCP Server

What profullstack/mcp-server does

profullstack/mcp-server is a modular MCP application that exposes a broad collection of developer and data-access modules through one server. Its model layer supports text generation, speech-to-text, image generation, and streaming inference through OpenAI, Anthropic, Stability AI, and Hugging Face integrations.

The utility modules cover several distinct workflows. SEO ranking uses ValueSERP, WCAG testing uses Pa11y, tides and currents data comes from NOAA CO-OPS, and Data.gov search provides cursor-paginated catalog results. Other modules search Craigslist, shorten links through hynt.us, create crypto payment badges, generate fake JSON with OpenAI, scan web applications, and verify webhook signatures for Stripe, GitHub, Slack, Shopify, Twilio, or generic HMAC payloads.

The fleet tools are specific to Profullstack services: they list products and surfaces, retrieve service details, search by capability, and explain CoinPay OAuth authentication. Social modules provide posting, login, status, platform information, content validation, sample content, statistics, and module metadata through @profullstack/social-poster.

How it works

The project uses a dynamically loaded module architecture. Modules can have their own packages and metadata, while the core server provides routing, model management, logging, status endpoints, health checks, and metrics. Model workflows select or activate a provider model, then send inference requests to the active or explicitly named model.

The application uses ES modules throughout. It starts an HTTP service on port 3000 by default and includes Docker and Docker Compose support. A hosted deployment is available through Fronteir AI.

Setup and configuration

Run the application with Node.js 18 or newer and pnpm 10 or newer:

pnpm install
pnpm start

Use pnpm dev for development with automatic reloads. Configuration is loaded from .env; the project includes a sample environment file. Add provider keys for the model features you intend to use, such as OPENAI_API_KEY, STABILITY_API_KEY, and ANTHROPIC_API_KEY. Hugging Face model support is also listed among the integrations.

For containerized deployments, build the image and publish port 3000:

docker build -t mcp-server .
docker run -p 3000:3000 mcp-server

Tools and capabilities

• Generate and update README shields.io badges using protected, contained paths.
• Search Craigslist, check SEO rankings, run WCAG tests, and scan web applications.
• Post to supported social platforms and inspect posting status, capabilities, and statistics.
• Search Datamart, retrieve resources, inspect provider capabilities, and find government services.
• Search Data.gov datasets and query NOAA marine navigation data.
• Verify signed webhooks with timing-safe, replay-resistant checks.
• Generate crypto badges, shorten links, and create OpenAI-backed fake JSON responses.
• Manage models and run text, image, and speech-to-text inference.

Limitations and notes

There is no global authentication layer. Any client that can reach the server port can call module routes unless a module-specific control applies, so deployments beyond localhost should sit behind an authenticating reverse proxy.

Scanner and README badge routes accept bearer or API-key tokens when their corresponding token variables are configured. If those variables are unset, the routes remain open and the server logs a warning. The server also restricts selected outbound requests to public addresses and applies CSRF checks to state-changing requests with foreign Origin headers. Browser origins can be configured with CORS_ORIGINS; clients that omit Origin, including command-line and server-to-server clients, are unaffected.