Agent Template

Build Deep Agents with LangGraph, config-as-code orchestration, MCP integration, and enterprise observability on OpenShift.

The Agent Template is a production-ready foundation for building Deep Agents with LangGraph via the Aegra CLI. It provides orchestrator and subagent patterns, MCP tool integration, conversation persistence, and deployment overlays for OpenShift and Kind.

Overview

This template gives you a complete Deep Agent runtime:

• Orchestrator + subagents — delegate work across specialized agents (example domain: fitness assistant)
• Skills — reusable workflow documents (client-intake, bmi-report, email-formatter) following the Agent Skills specification
• Config-as-code — prompts, subagents, MCP wiring, and runtime settings in config/agent/
• MCP integration — SSO pass-through, OAuth, and Dynamic Client Registration (DCR)
• Enterprise infrastructure — PostgreSQL checkpoints, Redis SSE broker, Langfuse tracing, OpenTelemetry metrics

MCP servers and the chat UI are separate repositories. This template runs the agent and its dependencies (Postgres, Redis) only.

Key Features

Deep Agent orchestration The orchestrator routes requests to subagents, manages task queues, and coordinates multi-step workflows with human-in-the-loop (HITL) support.

MCP tool integration Register MCP servers in mcp.json and attach them to the orchestrator or individual subagents via markdown frontmatter. Multiple auth modes for enterprise deployments.

Config-as-code runtime Customize agent behavior by editing PROMPT.md, subagents/*.md, and YAML config — no Python changes for most operational updates. See Architecture for details.

Production deployment Red Hat UBI container, OpenShift overlays with HPA and PDB, Kind overlay for local Kubernetes testing. Health endpoints at /health, /readyz, and /livez on port 5002.

Architecture

The diagram below illustrates the example domain shipped with the template—a Red Hat Fitness Assistant—not a generic blank agent. It shows how the runtime components connect when you run the repo out of the box.

In this example, a client (template-ui or any LangGraph API client) sends requests to the Aegra API on port 5002. The orchestrator handles intake and routing, then delegates to two subagents:

• Analyst — runs health-metric analysis and research using MCP tools (for example BMI calculation)
• Publisher — formats output and handles delivery tasks such as email

Tools are provided by template-mcp-server on port 5001. Prompts, subagent definitions, and MCP wiring live in config/agent/ and load at startup—no embedded Python prompts.

When you adapt the template for your own domain, you replace the fitness-assistant prompts, subagent names, and skills while keeping the same architecture: orchestrator, subagents, MCP client, PostgreSQL, and Redis.

Figure 1. Example Deep Agent request flow for the fitness-assistant template: orchestrator delegates to analyst and publisher subagents, which call MCP tools on template-mcp-server. Postgres stores checkpoints and feedback; Redis backs SSE streaming and OAuth tokens.

Getting Started

git clone https://github.com/redhat-data-and-ai/template-agent.git
cd template-agent
make install
make local
curl http://localhost:5002/health

Follow the Quick Start for credentials, MCP setup, and API testing.

Use Cases

Multi-agent workflows Orchestrate complex tasks across specialized subagents with clear delegation boundaries.

Enterprise tool access Connect to internal systems via MCP servers with SSO, OAuth, or DCR authentication.

Configurable domain agents Adapt the example fitness assistant to your domain by replacing prompts, skills, and subagent definitions.

Observable production agents Trace conversations with Langfuse, export OpenTelemetry metrics, and collect user feedback via the API.

Technology Stack

• LangGraph Deep Agents — orchestration framework
• Aegra CLI — LangGraph Platform dev server
• Python 3.13+ — with uv package manager
• PostgreSQL — checkpoints, memory, feedback storage
• Redis — SSE broker and OAuth token storage
• pytest — unit, integration, and skills evaluations

Related Templates

• MCP Server Template — tools for your agent to call
• UI Template — chat interface for the LangGraph API
• Stack reference — ports, request flow, deployment patterns
• Config-as-code announcement — how configuration is organized

Next Steps

1. Quick Start — run locally in minutes
2. Architecture — config layout, MCP auth, API reference
3. Deployment Guide — OpenShift, containers, production hardening
4. GitHub repository — latest README and issues