Mastra
FreemiumThe leading TypeScript framework for agents that run for days.
Published 27 September 2026
Scores
Popularity4/5
Over 27K GitHub stars and roughly 1M weekly npm downloads with adoption at Replit, Salesforce, and WorkOS mark it as the clear leader among TypeScript-first agent frameworks, though still smaller than the largest Python frameworks.
Learning Curve2/5
Typed agent and workflow configuration reads like idiomatic TypeScript, and JS/TS developers already familiar with Node.js tooling can get an agent running quickly without adopting a new mental model.
Flexibility4/5
Typed workflows with branching and retries, pluggable model routing, and MCP support cover most agent architectures, though the framework stays opinionated toward its own agent/workflow primitives rather than being fully unopinionated.
Performance4/5
Built-in evals, tracing, and metered CPU execution on Mastra Cloud are designed for production workloads that run continuously, backed by production deployments processing thousands of daily agent runs.
Portability4/5
Apache 2.0-licensed core is self-hostable anywhere Node.js runs and is provider-agnostic across 40+ model providers, though the managed Mastra Cloud layer and enterprise features introduce some pull toward the vendor's own platform.
About Mastra
Mastra is an open-source TypeScript framework for building AI agents, workflows, and retrieval-augmented applications, built by the team behind the Gatsby framework. It positions itself as the JavaScript-first alternative to Python-centric agent frameworks like LangChain and LlamaIndex, integrating directly into React, Next.js, and Node.js codebases or running as a standalone server, rather than bolting an AI layer onto an existing JS app from the outside.
Agents are defined with a typed configuration covering instructions, model, tools, and runtime behavior, so agent definitions get the same compile-time safety as the rest of a TypeScript codebase. Workflows compose typed steps with retries and branching for multi-stage automation, and a built-in memory system gives agents observational memory, semantic recall, and thread-aware storage that persists context across long-running sessions. Model access is standardized through routing across 40+ providers, including OpenAI, Anthropic, and Gemini, and agents can be exposed as Model Context Protocol (MCP) servers or consume external MCP tools.
Mastra ships with production-facing tooling out of the box: evals, metrics, traces, and datasets for monitoring agent quality and behavior over time, rather than treating observability as a separate integration. The core framework is Apache 2.0-licensed and free to self-host with no seat or usage limits; a managed Mastra Cloud platform adds hosted observability, CPU-metered execution, and enterprise controls (SSO, RBAC, VPC data isolation) on top, priced separately from the open-source library. Production users include Replit, Salesforce, Sanity, SoftBank, and WorkOS.
Key Features
- Typed agent definitions: instructions, model, tools, and runtime behavior in one config
- Workflow engine: typed steps, retries, and branching for multi-stage automation
- Observational memory with semantic recall and thread-aware durable context
- Standardized model routing across 40+ providers including OpenAI, Anthropic, and Gemini
- Native Model Context Protocol (MCP) server authoring and consumption
- Built-in evals, metrics, traces, and datasets for production observability
Pros
- TypeScript-native design fits directly into existing React/Next.js/Node.js codebases
- Apache 2.0-licensed core, free to self-host with no usage limits
- Built-in observability (evals, traces, metrics) instead of a bolt-on integration
- Backed by an experienced team (ex-Gatsby) and a well-funded YC-backed company
Cons
- TypeScript/JavaScript-only, no path for Python-first teams
- Younger ecosystem than LangChain/LlamaIndex, with fewer third-party integrations
- Mastra Cloud pricing and enterprise features are separate from the open-source core
- Smaller community and job-market footprint than the largest Python frameworks
Mastra Pricing
Freemium- · Apache 2.0-licensed framework, self-hosted anywhere Node.js runs
- · Full agents, workflows, memory, and MCP support with no usage limits
- · 100K observability events and 24 CPU hours included, then metered usage
- · 15-day data retention, unlimited users, deployments, and projects
- · 1M observability events and 250 CPU hours included, then metered usage
- · 6-month data retention, multiple teams, SSO, and SOC 2 documentation
- · Custom volume, retention, RBAC, audit logs, uptime SLAs, and dedicated support
- · VPC data isolation and flat annual pricing with no per-trace metering
Tools Related to Mastra
Works well with Mastra(7)
Next.js is Mastra's most common deployment target and has a dedicated official getting-started guide — agents and workflows typically run as Next.js API routes or server actions.
Mastra has an official SvelteKit getting-started guide for wiring agent and workflow calls into a SvelteKit app.
Mastra has an official React + Vite getting-started guide for wiring agent and workflow calls into a React front end via the Mastra Client SDK.
Mastra has an official Nuxt getting-started guide, covering the Vue ecosystem the same way its Next.js guide covers React.
Mastra's TypeScript-native design and Client SDK work with plain Vue apps, though the official framework guide targets Nuxt rather than bare Vue.
Mastra ships a first-party @mastra/nestjs server adapter — a dedicated NestJS module that registers Mastra's agent and workflow routes inside an existing NestJS application.
Integrates with Mastra(3)
Mastra ships a first-party OpenAI provider as part of its standardized model routing layer, for running agents on OpenAI models.
Mastra ships a first-party Anthropic provider as part of its standardized model routing layer, for running agents on Claude models.
Mastra ships a first-party Google provider as part of its standardized model routing layer, for running agents on Gemini models.
Alternatives to Mastra(5)
The closest head-to-head comparison for TypeScript teams — the Vercel AI SDK is a leaner provider-agnostic toolkit for streaming and tool calls, while Mastra adds a fuller framework layer of workflows, memory, and evals on top of the same JS-native audience.
Both are agent frameworks for building multi-step, tool-using agents — CrewAI centres on role-based multi-agent crews in Python, Mastra on typed agents and workflows in TypeScript. Teams pick one as their primary framework.
Both frameworks let developers hand-build multi-step agent workflows with explicit control flow — LangGraph as a Python state-graph runtime, Mastra as a typed TypeScript workflow engine.
Mastra positions itself as the TypeScript-first alternative to LangChain, covering the same RAG, memory, and multi-agent workflow ground with typed, JS-native primitives instead of LangChain's Python-first component library.
Both are type-safe, production-oriented agent frameworks built around strict typing — Pydantic AI for Python teams, Mastra for TypeScript teams. Competing picks depending on backend language.
Learning Resources
No resources yet — check back soon.
Vendor
Mastra
Website →Tags
Details
- Maintained
- Yes
- Agent type
- Framework
- Primary language
- TypeScript
- Open source
- Yes
- Hosting
- Self-hosted
- GitHub stars
- 28.3k
- Stars updated
- 2026-09-23