Mastra

What it is for

Mastra is an Apache-2.0 TypeScript framework for agents: npm create mastra@latest, attach tools, add memory and workflows, then run locally or on Mastra Platform. Use it when the job is ship an agent in Node/Next, not when you want a hosted chat seat with no repo.

What makes it unique: the cover is the official mastra-ai/mastra GitHub file list — .agents/skills, agent-sdks, Apache-2.0 — not a wordmark.

Key features

• Agents + tools — Agent + createTool so the model calls your systems, then agent.generate() (docs). MCP / model-router strings are the same act stream.
• Workflows — typed steps with .then() / .branch() / .parallel(), retries, and pause-for-human (homepage). Control flow, not a chat.
• Memory — observational memory, semantic recall, and thread storage so the agent keeps useful context (Memory). Remembering, not another generate.
• Evals + datasets — code or model judges plus versioned test cases so you compare a change before you ship (Observability). A score, not a reply.
• Tracing — follow model calls, tool calls, and workflow steps in a run (same Observability docs). Debug, not a second agent.

Studio is the local UI for this project. Factory / Harness / create mastra sit on the same agent runtime. Platform events, CPU hours, and Gateway markup are billing. MCP authoring is SDK.

Who it's for

Best fit: TypeScript teams who will run Node and already have provider keys.

Pricing

Verified 2026-10-22 (Asia/Shanghai) from mastra.ai/pricing and the homepage Apache-2.0 note.

Software (self-host)

• Core framework — Apache-2.0. License fee $0. You bring model keys.
• Self-hosted Enterprise — contact sales. Flat annual fee; data stays in your VPC. No public dollar.

Mastra Platform (hosted)

• Starter — $0/month. 100K observability events (then $10 / 100K); 24 CPU hours (then $0.35/hr); 15-day retention; unlimited users / deployments / projects.
• Teams — $250/month. 1M events (then $8 / 100K); 250 CPU hours (then $0.25/hr); 6-month retention; multiple teams, SSO, SOC 2 docs.
• Enterprise — custom volume / retention, RBAC, audit logs, SLAs.
• Persistent server (24/7) — $100/project on Starter and Teams (custom on Enterprise).
• Gateway tokens — market rate + 5.5% on all Platform cards we copied.

Limitations / costs

• Framework $0 is not tokens. Platform Starter is $0 until you blow 100K events or 24 CPU hours.
• Teams is $250/month, not a seat card.
• Library + Studio, not a hosted GTM inbox. For that, see Gumloop or Lindy.

Alternatives

• Pydantic AI — Python agent SDK with a public license.
• Agno — Python agent framework.
• Letta — stateful memory harness (Apache-2.0 + hosted Pro $20).

Official tutorial

Official Mastra channel video: Build and deploy your first agent with Mastra.

Written path: Get started (npm create mastra@latest) plus Pricing.

Website

https://mastra.ai

Repository

https://github.com/mastra-ai/mastra

Source

Catalog row curated for AI Tools Nav. Cover is the official mastra-ai/mastra GitHub file list. Dollars checked on mastra.ai/pricing (framework Apache-2.0 $0; Platform Starter $0; Teams $250/mo; persistent server $100/project). Tutorial oEmbed author: Mastra.

Last verified

2026-10-22 (Asia/Shanghai) editorial check.