AgentStack

AgentStack is a production-grade multi-agent framework built on Mastra that coordinates a large fleet of agents, tools, and workflows to enable scalable AI-powered applications. The MCP orchestration layer allows agents to run in parallel, share context, and pass information through workflows, delivering end-to-end automation for tasks such as financial analysis, RAG pipelines, and governance. With 41 specialized agents, 21 workflows, and 11 networks, you can compose complex AI-driven processes that are observable, secure, and maintainable. Use AgentStack to deploy ready-made capabilities like financial intelligence, document processing, and cross-agent collaboration, or extend it with your own agents and tools to fit enterprise needs. The included UI components and observability tooling (Langfuse traces, 10+ custom scorers, and robust RBAC) help you monitor performance, enforce security, and maintain high-quality results across your AI stack.

To use AgentStack effectively, start by launching the MCP server and connecting your agent tools. The server coordinates parallel execution across agents, enabling scenarios such as stock analysis, report generation, and governance workflows. Explore the available tools and networks to see how agents interact, and leverage the Zod-based schemas to validate data boundaries and ensure type safety throughout the system. You can customize the orchestration by configuring tools, workflows, and roles, then deploy additional agents as your use case expands.

Prerequisites:

• Node.js >= 20.9
• npm or pnpm
• Git
• Optional: a running Postgres database and Langfuse account for observability

0. Clone the repository

git clone https://github.com/ssdeanx/AgentStack.git
cd AgentStack

1. Install dependencies

npm install
# or if you prefer pnpm:
pnpm install

2. Configure environment

• Create a copy of the example env and populate required values

cp .env.example .env

Edit .env to include at minimum:

• NODE_ENV=production
• LANGFUSE_API_KEY=your-langfuse-api-key
• JWT_SECRET=your-jwt-secret
• DATABASE_URL=postgresql://user:pass@host:port/dbname

3. Build (if applicable) and start the server

npm run build
npm run start

If your environment uses a different startup script, ensure the entry point is reachable (see mcp_config in this document).

4. Verify operation

• Check logs for MCP orchestration activity
• Ensure agents are registering and communicating via MCP
• Confirm observability dashboards (Langfuse) are receiving traces

Tips and common issues:

• Environment variables: Ensure LANGFUSE_API_KEY, JWT_SECRET, and DATABASE_URL are correctly set to enable tracing, authentication, and data storage.
• If you migrate to a different database or host, update DATABASE_URL accordingly and run migrations if required by your setup.
• For debugging, run in development mode (NODE_ENV=development) to enable verbose logs, then switch to production for deployment.
• The MCP server coordinates parallel agents; if you see bottlenecks, review workflow definitions and increase concurrency where safe.
• Ensure your Node.js version matches the minimum requirement (>= 20.9) to avoid runtime incompatibilities with modern dependencies.
• If you adapt the stack with additional tools or agents, maintain the Zod schemas for data validation to preserve end-to-end type safety.