Deployment & Runtime

These packages give you a FaaS runtime where functions are exposed as MCP tools, letting AI agents call them directly without cold starts. You'd adopt them to eliminate the latency of cold-start function invocations while making your serverless functions natively callable by AI agents through the Model Context Protocol. The most distinctive thing is that functions are MCP tools first and HTTP endpoints second, with a warm pod pool manager that keeps pre-warmed containers ready for sub-100ms invocations.

These packages let you run multiple Node.js serverless functions inside a single Docker container, with each function route automatically exposed as an MCP tool for AI agents. You'd adopt them to get the ergonomics of serverless—per-function routing, environment isolation, independent deployments—without cold starts, vendor lock-in, or per-request billing. The key architectural choice is a custom Map-based dispatcher that avoids Express router stack bloat, giving O(1) route lookup and clean hot reloads without middleware leaking across function updates.

These packages give you a documentation site generator that builds static HTML from Markdown files, with built-in support for versioned docs and search. You would adopt them to create and maintain technical documentation for your own open-source project or product without running a separate CMS or wiki. The most distinctive thing is that the core rendering pipeline is split into a Fastify plugin for development previews and a CLI for production builds, letting you reuse the same content processing logic in both server-side and static contexts.

These ~700 MIT-licensed TypeScript packages give you the building blocks to assemble voice agents, RAG pipelines, evals, security controls, observability, and deployment infrastructure for small business AI systems. You'd adopt them to stand up a virtual receptionist, knowledge agent, or customer-intake bot in an afternoon without getting locked into a proprietary platform or paying enterprise rates. The packages are designed to compose like puzzle pieces—each one is a focused, production-ready component (e.g., a Fastify plugin, an OpenTelemetry GenAI exporter, a Drizzle adapter) that you can pick up or leave out, and the whole collection ships with end-to-end solution recipes that are typecheck-clean and test-covered.

These Terraform modules deploy a Model Context Protocol (MCP) server workload onto Amazon EKS, provisioning the full supporting infrastructure: a cluster, ElastiCache Redis for session state, SQS task and result queues with a dead-letter queue, Secrets Manager secrets, and a CloudWatch dashboard with SNS alerts. You would adopt them to get a production-ready MCP deployment on AWS without wiring together EKS, KEDA, Redis, SQS, IRSA, and monitoring yourself. The modules are designed as composable building blocks—the root module wires everything together for a drop-in experience, while individual submodules let you swap in your own KEDA installation or selectively assemble only the pieces you need.

These Terraform modules deploy any MCP (Model Context Protocol) server onto Google Cloud Run, wiring it to Firestore for session state, Secret Manager for credentials, and Pub/Sub for async task distribution. You'd use them to get a production-ready MCP server infrastructure on GCP without designing the networking, IAM, observability, and persistence plumbing yourself. The modules are designed as drop-in components that compose through Terraform outputs and service account emails, letting you mix and match Cloud Run, Firestore, Secret Manager, and Pub/Sub modules independently for different environments or service topologies.

This Terraform module deploys a complete observability stack—traces, metrics, logs, and alerting—for MCP-based agent systems onto AWS. You would adopt it to get a pre-wired, one-command observability setup instead of manually configuring and connecting OpenTelemetry collectors, Prometheus, Grafana, CloudWatch, and a trace backend. Its most distinctive trait is that the entire stack is configured as a single Terraform module with selectable trace backends (Arize Phoenix or Langfuse), pre-built GenAI dashboards, and PromQL alerting rules, all consuming OpenTelemetry output from your instrumented agents.