Mastra
by Mastra (founded 2024 by Sam Bhagwat, Abhi Aiyer and Shane Thomas, the Gatsby founders; Y Combinator-backed; San Francisco, ~45 employees remote). Funding: USD 13M seed (announced 2025-10-08) and USD 22M Series A led by Spark Capital (2026-04-09, per secondary reporting), USD 35M total per mastra.ai/about. Mastra 1.0 stable shipped 2026-01-21 (secondary source).
The TypeScript AI framework for building agents, workflows, and tools
Features
- Model Routing: Connect to 40+ LLM providers (OpenAI, Anthropic, Gemini, Llama) through one unified interface
- Agents: Build autonomous agents that use LLMs and tools to solve open-ended tasks with reasoning capabilities
- Workflows: Graph-based orchestration engine for complex multi-step processes with intuitive control flow
- RAG (Retrieval-Augmented Generation): Built-in knowledge integration and semantic recall
- MCP Support: Author Model Context Protocol servers to expose agents via standard interfaces
- Memory Management: Conversation history, working memory, and context management for coherent agent behavior
- Human-in-the-Loop: Suspend and resume execution with persistent state management
- Evals & Observability: Built-in scorers, logging, tracing, and evaluation tools for production monitoring
- Framework Integration: Works with Next.js, Express, Hono, React, and Node.js
- Deployment Flexibility: Ship as standalone APIs or bundle within existing applications
Superpowers
Mastra is designed for TypeScript developers who want to build production-ready AI applications without learning Python. The framework’s philosophy is “Python trains, TypeScript ships.”
Who this is for:
- TypeScript/JavaScript developers building AI agents and workflows
- Teams prototyping and scaling AI features in existing codebases
- Developers needing production-ready observability and evaluation tools
- Organizations wanting to deploy AI agents with proper monitoring and controls
What you gain:
- Rapid Development: Interactive dev studio with local server for agent iteration
- Production Ready: Built-in observability, token tracking, and identity-based access control
- Open Ecosystem: Apache 2.0 core license with ~28.5k GitHub stars (2026-09-30; was 19.1k in Jan 2026) and active Discord community
- Unified Interface: Work with any LLM provider without learning multiple APIs
- End-to-End Platform: From prototype to production with integrated studio, observability, and deployment tools
Plans
Current prices: see the vendor pricing page https://mastra.ai/pricing.
Mastra Framework (Open Source):
- Free - Apache 2.0 core;
ee/directories fall under the Mastra Enterprise License (source-available, licence needed for production) - Includes: Agents, workflows, RAG, evals, MCP servers, integrations
Mastra Platform (cloud, per mastra.ai/pricing, accessed 2026-10-02):
- Starter (free): 100K observability events, 24 CPU hours, 15-day retention; usage beyond the included amounts is metered
- Teams: 1M events, 250 CPU hours, 6-month retention; usage beyond the included amounts is metered
- Enterprise: custom (RBAC, audit logs, SLAs, dedicated engineer)
- LLM tokens billed at market rate + 5.5%; persistent storage/server features cost extra
- Includes: Observability (logging, tracing, evals), Studio (collaborative UI), Deployment
Custom Support & Deployment:
- Enterprise support with architecture meetings, shared Slack, SLAs
- On-premises deployments with SSO and RBAC
- Contact the vendor for enterprise terms
Getting Started
npm create mastra@latest Resources:
- Documentation: https://mastra.ai/docs
- GitHub: https://github.com/mastra-ai/mastra
- Community: Discord (link on website)
- Templates: Pre-built solutions for browser automation, data analysis, coding tasks
Integration Examples
- Vercel AI SDK: Build agentic UIs with streaming responses
- Next.js: Embed agents in full-stack applications
- Standalone Server: Deploy as independent API endpoints
- CopilotKit: Create AI copilot experiences
Status (2026-10-02)
Active and fast-moving: ~28.5k GitHub stars, ~2.9k forks; @mastra/core 1.72.0 released 2026-09-29 (GitHub releases page; includes MCP v2 (2026-07-28 spec) server architecture since 1.68.0). Comparison with CrewAI: mastra-vs-crewai-comparison, crewai.
Related
- crewai, mastra-vs-crewai-comparison
- copilotkit
- model-context-protocol
- ai-agentic-systems