Start here
Choose your platform.
You don't need to evaluate eight frameworks. You need to answer three questions — language, statefulness, and MCP depth — and the right platform becomes obvious. This page is the shortest path to a working setup, sourced from official SDK documentation and community deployment evidence.
This page contains affiliate links — if you sign up through them we may earn a commission at no cost to you. Picks are made before monetization is considered. How we research →
The three questions
1. What language are you shipping in? Python teams have the widest choice. TypeScript teams should look at Mastra first. Polyglot enterprise teams need Google ADK.
2. Do you need stateful, long-running workflows with checkpointing? If yes: LangGraph. If you are prototyping first: CrewAI. If you need OS-level computer use: Claude Agent SDK.
3. How important is validated, structured output? If your agent feeds downstream systems that break on malformed output: Pydantic AI. Otherwise, any Python framework handles prose output adequately.
The recommendation map
- Quick multi-agent prototype (Python): CrewAI
- Production stateful pipeline: LangGraph
- TypeScript team: Mastra
- OS automation or heavy MCP: Claude Agent SDK
- Voice or multi-LLM: OpenAI Agents SDK
- Type-critical regulated domains: Pydantic AI
- Polyglot enterprise: Google ADK
The full decision guide with comparison table and honest trade-offs is in Which AI Agent Platform Should You Use in 2026?
Common questions
What does it cost to run an AI agent?
It depends entirely on the architecture. A simple stateless research agent using the Claude Agent SDK costs under $1/day for typical usage. A parallel multi-subagent research workflow can cost $5-20/session. Managed platforms (Claude Managed Agents, OpenAI Assistants API) cost more per call but require less infrastructure. Set max_budget_usd cost controls before deploying any agent to production.
Do I need to be a machine learning engineer to use these frameworks?
No. The frameworks on this site are designed for software engineers and technical professionals who can write Python or TypeScript. You are configuring agent behavior through prompts and SDK parameters — you are not training models. The audience for most build guides here ranges from "I can follow a tutorial" to "I ship production Python."
What is MCP and why does it matter?
Model Context Protocol is the standard that lets agents connect to external tools and data sources. Without MCP, giving an agent access to Gmail, a database, or a browser requires custom integration code. With MCP, you add a JSON config entry pointing to a community-maintained MCP server. There are 19,000+ servers on MCP.so as of July 2026. The MCP and Skills Ecosystem guide covers every registry in detail.
The Stack Brief
One sharp briefing, every week.
Platform releases, build guides, and trending repos -- sourced, verified, no vendor puffery. Free.
You're in. Watch your inbox. ✓