Build AI agents on Catalyst

This section is for developers who build AI agents and MCP servers. It takes you from your first agent to running agents in production: durable execution so an agent survives crashes and restarts, an identity for every agent, and policies that control which tools and services it can call.

You keep your agent framework. Catalyst works with Dapr Agents, LangGraph, CrewAI, Strands, OpenAI Agents, Google ADK, Pydantic AI, and more.

Start with your AI coding agent

Let Claude Code, Codex or the Copilot CLI run a durable agent in your framework for you.

Quickstart: Agent Durability (AI Plugin)

Your agent runs a sample agent, crashes it, and shows you it resume.

Local installation

Install the Catalyst AI plugin for your coding agent.

Pick your framework

Each framework has a page with its quickstart, tutorials, and access policies.

• Dapr Agents

  • Quickstart: CLI
  • Durable agent
  • All resources

• CrewAI

  • Quickstart: AI / CLI
  • Access policies

• LangGraph

  • Quickstart: AI / CLI
  • Access policies

• Strands

  • Quickstart: AI / CLI
  • Access policies

• Microsoft Agent Framework

  • Quickstart: AI / CLI
  • Access policies

• Spring AI

  • Quickstart: AI / CLI
  • Crash recovery
  • All resources

• Google ADK

  • Quickstart: AI / CLI
  • Access policies

• OpenAI Agents

  • Quickstart: AI / CLI
  • Access policies

• Pydantic AI

  • Quickstart: AI / CLI
  • Access policies

• Deep Agents

  • Quickstart: AI / CLI
  • Sub-agents quickstart
  • All resources

• Claude Agents

  • Quickstart: AI / CLI
  • Access policies

• HolmesGPT

  • Quickstart: CLI
  • Access policies

• Mastra

  • Quickstart: AI / CLI
  • Access policies

• Go AI

  • Durable Go agents
  • Access policies

For every framework

MCP servers

Connect agents to MCP servers through Catalyst, with an access policy for each tool.

Enterprise identity

Let agents act for a signed-in user, with tokens from your identity provider.

Multi-agent systems

Coordinate several agents with pub/sub and durable workflows.

Deploy to Kubernetes

Run your agents on your own Kubernetes cluster, connected to Catalyst.

Where to go next

Concepts

How agents, MCP, durable execution, and identities fit together.

Run & operate

Manage agents, MCP servers, and token budgets, and observe them in production.

Reference

The MCP server reference and the control plane API.

Learn hands-on

🎓Durable Execution & Dapr University · Learn this hands-onDapr Agents: Build LLM-powered intelligent applications with tools and workflowsBuild intelligent agents with custom tool calls, MCP integration, and multi-agent collaboration.Intermediate45 minPythonStart the track → 🎓Durable Execution & Dapr University · Learn this hands-onDapr Agents - Fix Outdated Agent AnswersGive agents live web results with Tavily so they stop answering from stale training data.Intermediate30 minPythonStart the track → 🎓Durable Execution & Dapr University · Learn this hands-onMake MAF agents reliable with Dapr WorkflowAdd durable state and crash recovery to Microsoft Agent Framework agents with Dapr Workflow.Intermediate30 min.NETStart the track → 🎓Durable Execution & Dapr University · Learn this hands-onMake LangGraph agents durable with Dapr Workflow: Schedule PlannerCrash a LangGraph agent mid-run and watch Dapr Workflow resume it where it stopped. No API key needed.Intermediate30 minPythonStart the track → 🎓Durable Execution & Dapr University · Learn this hands-onMake LangGraph agents durable with Dapr Workflow: Supply Chain AuditorRun an agent that audits a dependency bump for supply-chain attacks, crash it, and resume without repeating the LLM call.Intermediate30 minPythonStart the track → 🎓Durable Execution & Dapr University · Learn this hands-onMake DeepAgents durable with Dapr Workflow: Deep Issue InvestigationCrash a DeepAgents investigation mid-run and resume it from durable state without re-running finished LLM calls.Intermediate30 minPythonStart the track →