Mastra — TypeScript AI Agent Framework

What it is

Mastra is a production-ready TypeScript framework for building AI agents. It provides first-class support for tool use, workflows, retrieval-augmented generation (RAG), and agent memory. Mastra integrates natively with the Model Context Protocol (MCP), letting agents connect to external services through standardized tool interfaces.

TypeScript developers building AI-powered applications who want type safety, IDE autocompletion, and a familiar Node.js deployment model will find Mastra a natural fit.

How it saves time or tokens

Mastra eliminates the need to glue together separate libraries for LLM calls, tool schemas, vector search, and workflow orchestration. Everything ships in one framework with TypeScript types. The estimated token usage per session is approximately 2,800 tokens.

How to use

1. Install Mastra via npm.
2. Define agents with tools, instructions, and model configuration.
3. Deploy as a standard Node.js application.

import { Agent } from '@mastra/core';
import { openai } from '@mastra/openai';

const agent = new Agent({
  name: 'researcher',
  model: openai('gpt-4o'),
  instructions: 'You are a research assistant. Search and summarize findings.',
  tools: [searchTool, summarizeTool],
});

const result = await agent.generate('Find recent papers on prompt compression');
console.log(result.text);

Example

A multi-step workflow that researches a topic, drafts a summary, and sends it via email:

import { Workflow } from '@mastra/core';

const researchWorkflow = new Workflow({
  name: 'research-and-report',
  steps: [
    { id: 'research', agent: researchAgent, input: '{{topic}}' },
    { id: 'draft', agent: writerAgent, input: '{{research.output}}' },
    { id: 'send', tool: emailTool, input: { body: '{{draft.output}}' } },
  ],
});

Related on TokRepo

• Multi-agent frameworks — Compare Mastra with other agent orchestration tools
• AI tools for agents — Browse agent frameworks and toolkits

Common pitfalls

• Mastra is TypeScript-only. Python-based teams should look at LangChain or CrewAI instead.
• MCP server compatibility depends on the server implementation. Test integrations individually.
• Workflow error handling requires explicit try-catch or error step definitions; there is no automatic retry by default.