Mastra

Mastra is an open-source TypeScript framework for building production-ready AI agents, agentic workflows, and RAG pipelines. Created by the team behind Gatsby (Sam Bhagwat CEO, Abhi Aiyer CTO, Shane Thomas CPO), it graduated from Y Combinator's Winter 2025 batch and announced a $13M seed round. The framework reached v1.0 in January 2026 and has 27.6k GitHub stars. Mastra fills a gap in the agent ecosystem: LangChain, CrewAI, and AutoGen began in Python with later JavaScript ports, while Mastra was designed from the ground up for TypeScript. The framework provides core primitives for agent development. Agents are autonomous LLM-driven decision makers. Tools provide typed external integrations via Zod schemas. Workflows enable deterministic multi-step pipelines. Additional built-in capabilities include memory systems, RAG with vector database management, supervisor patterns for multi-agent orchestration, evals, observability, and Mastra Studio (a visual IDE for testing and debugging). The model router connects to models from 40+ providers (OpenAI, Anthropic, Google Gemini, DeepSeek, Groq, Mistral, xAI Grok, Amazon Bedrock, Azure OpenAI, Ollama, OpenRouter, Vercel, Hugging Face, Cohere, Cerebras, and more) through one standard interface with automatic fallbacks and IDE autocomplete. Mastra covers both directions of the Model Context Protocol. MCPClient connects a Mastra agent out to external MCP servers, and MCPServer exposes Mastra agents, tools, workflows, prompts, and resources to other MCP-compatible systems. Both are documented in the vendor's own reference documentation and ship in the @mastra/mcp package. Production users with published case studies include Replit (Mastra powers Replit Agent's agent builder), Salesforce, Sanity, SoftBank, Factorial, WorkOS, Plaid, Cedar, Counsel Health, and Medusa. Pricing: the core framework is open source under Apache 2.0 and free for any use. Enterprise features (RBAC, SSO, ACL) reside in the ee/ directory and require a commercial license for production use but are freely usable for development and testing. Mastra Cloud is a hosted platform with GitHub-connected deployments, autoscaling, instant rollbacks, and centralized observability. Starter is free at $0/month (100K observability events, 24 CPU hours, 15 days data retention). Teams is $250/month (1M events, 250 CPU hours, 6 months retention, multiple teams, SSO, SOC 2 documentation). Enterprise is custom pricing on a sales call (custom volume and retention, RBAC, audit logs, support and uptime SLAs, dedicated support engineer). Self-hosted deployment is available at no cost, with a separately licensed Enterprise tier for RBAC, SSO, IAM and network policy integration. All published rates are USD. What the entry price does not buy: usage beyond each plan's included allowance is billed on top, at $10 per 100K observability events and $0.35 per CPU hour on Starter, and $8 per 100K events and $0.25 per CPU hour on Teams. There is no annual billing option published on either surface. Best fit: TypeScript and JavaScript developers building production AI agents without learning Python, full-stack web teams adding agentic features to Next.js, React, Express, or Hono apps, and platform companies (like Replit) embedding agent infrastructure into their own products. Who Mastra is not for: Python-only shops or teams committed to LangChain and CrewAI Python workflows cannot use Mastra without rewriting in TypeScript. Teams needing turn-key no-code agent builders should consider Botpress, Voiceflow, or Microsoft Copilot Studio instead. Teams handling regulated data should read the terms first, because Mastra's agreement prohibits submitting protected health information under HIPAA, cardholder data under PCI DSS, and data covered by the Gramm-Leach-Bliley Act unless separately agreed in an order form. Current state Q3 2026: Mastra holds a SOC 2 Type II report received in October 2025, published on its trust center at trust.mastra.ai, and it is the only certification the vendor publishes. Its terms of service carry an explicit no-training clause: Mastra does not use customer inputs or outputs to train or fine-tune any foundation model, except on express opt-in instruction or as aggregated de-identified usage data. The @mastra/core package recorded 1.2 million npm downloads in the week of 9 August 2026. The G2 profile is live but unclaimed and carries no reviews.