Mastra
A TypeScript framework for building, orchestrating, evaluating, and deploying AI applications and agents.
- Source repo
- mastra-ai/mastra
- Stars
- ★ 29k
- Last updated
- today
- License
- NOASSERTION
- Primary language
- TypeScript
- FA score
- 52/100 · Major gaps
At a glance
- How it runs
- Works with
- Universal · cross-platformOpenAI API · Claude API
- Cost
- Free software; you pay for model usage
- Setup effort
- Low · running in minutes
- You'll need
- Typical use
- A TypeScript team adding a tool-using customer or internal assistant with conversation history to a Next.js product.
- Not a fit if
- Teams that want to avoid any enterprise-licensed code paths
- Teams building on non-TypeScript stacks
- Source review
- 52/100 · Major gaps
What does this agent do, and when should you use it?
Mastra is a modern TypeScript framework for AI-powered applications and agents. It combines model routing, agents, graph-based workflows, context and memory features, plus evaluation and observability capabilities. Its agents use LLMs and tools for open-ended work, iterating internally until a final answer or optional stopping condition; workflows provide explicit execution control when that is preferable. Teams can integrate it into React, Next.js, or Node applications, deploy it as a standalone server, and author MCP servers that expose agents, tools, and structured resources. It fits teams that want to move from prototype to production on a TypeScript stack while retaining provider choice and orchestration control.
Developers create a project with npm create mastra@latest and select openai, anthropic, google, or xai as the LLM provider during creation. Mastra's model router connects to more than 40 providers through one interface; agents can use conversation history, retrieve data from APIs, databases, and files, and use tools while working toward a final answer. For deterministic multi-step control, workflows define execution with .then(), .branch(), and .parallel(); executions can suspend for user input or approval and resume using state remembered through storage. The framework can also expose agents, tools, and structured resources through an MCP server, or package them into web applications and standalone endpoints.
- A TypeScript team adding a tool-using customer or internal assistant with conversation history to a Next.js product.
- A developer connecting API, database, and file retrieval to an LLM-driven process that may iterate before returning an answer.
- An engineering team that needs explicit branching, parallel work, and step sequencing instead of leaving process control entirely to an agent.
- A business application that must pause an execution for a human approval or missing user input, then resume it later.
- A developer exposing custom agents, tools, or structured resources to systems that support MCP.
How do you install or deploy this agent?
The documented recommended project-creation command is npm create mastra@latest. The creator asks for a provider; the documented non-interactive form is npm create mastra@latest <project-name> -- --llm <provider>, where <provider> must be openai, anthropic, google, or xai. The supplied material does not document provider credential configuration.
How do you use this agent?
After creation, enter the project directory and run npx bgproc start -n <project-name> -w -- npm run dev. Open Mastra Studio at http://localhost:4111 to build, test, and manage agents, workflows, and tools. Integrate the resulting application with React, Next.js, or Node, or deploy it as a standalone server.
What are this agent's strengths and limitations?
- Model routing offers one interface to more than 40 providers and explicitly names OpenAI, Anthropic, and Gemini.
- It supports both autonomous agents and graph-based workflows with
.then(),.branch(), and.parallel()control flow. - Suspension and resumption are documented with execution state retained through storage, supporting human approval points.
- It includes evals and observability and can author MCP servers for agents, tools, and structured resources.
- Project creation requires choosing from a constrained provider list, while the supplied material does not explain credentials, secret handling, or local environment variables.
- Features in directories named
ee/use the Mastra Enterprise License and require a valid enterprise license for production use. - The supplied material does not state minimum Node.js, npm, database, or deployment-environment versions, so teams need to validate their runtime separately.
- Vercel AI SDK UI and CopilotKit integrations are mentioned, but no integration code or compatibility limits are included in the supplied material.
How does this agent compare with similar options?
Key facts side by side with the most closely related agents.
| Agent | Source review | Form / cost | Stars | Updated | Language | Full support on |
|---|---|---|---|---|---|---|
| Mastra This agent | 52 · Major gaps | CLIFree + model costs | ★ 29k | today | TypeScript | OpenAI API · Claude API |
| AgentScope 2.0 | 70 · Some gaps | Library / SDKFree + model costs | ★ 35k | 2d ago | Python | OpenAI API · Claude API |
| Pocket Flow | 41 · Major gaps | Library / SDKFree + model costs | ★ 11k | 3mo ago | Python | — |
| elizaOS | 88 · Good | CLIFree + model costs | ★ 20k | 1d ago | TypeScript | — |
How does FollowAgents rate this agent?
Why each dimension lost points
Evidence: README mentions human-in-the-loop, implying user confirmation exists, but no implementation details; data flow transparency is hinted via observability and evals, but not detailed; sensitive data handling not specified; dependency security has CodeQL workflow and lock files, but no vulnerability scan results; external effects may involve external API calls, but permission controls unclear; rollback not mentioned; source attribution clear with copyright and license, but publisher unverified. Deductions: most trust criteria only surface-level mentions without concrete implementation evidence.
Evidence: clear project structure, extensive test scripts and CI configs indicate good self-consistency; dependency management uses pnpm and catalogs, but no availability guarantees; failure messages not clearly documented. Deductions: failure messages criterion lacks concrete evidence.
Evidence: README clearly targets TypeScript developers, offers multiple use cases (agents, workflows, integrations); capability boundaries documented but not exhaustive; trigger precision mentioned via stopping conditions, but not detailed; environment fit supports multiple frameworks and deployment options. Deductions: trigger precision and some capability boundaries not specific enough.
Evidence: clear information architecture with docs, reference, guides; detailed installation instructions via CLI and manual; stable naming with version numbers; rich examples and FAQ; known limitations not explicitly listed; license clear with dual-license; versioning changelog exists (changesets); maintenance responsibility clear with contribution guide. Deductions: known limitations not explicitly listed.
Evidence: output usability via standard interfaces and tools, but no concrete output examples; marginal value via multiple features, but no comparison with other frameworks; cost-benefit open-source free, but no performance data. Deductions: lack of concrete data.
Evidence: README claims supported by documentation links, but no implementation code; cross-source corroboration via multiple doc pages, but no independent verification; fact-inference separation distinguishes features and recommendations, but not explicitly labeled. Deductions: cross-source corroboration and fact-inference separation insufficient.
- Publisher identity unverified; be cautious about supply chain risks.
- Dependency security only has CodeQL workflow, no vulnerability scan results; check dependencies for vulnerabilities.
- Sensitive data handling and permission control details insufficient; evaluate before use.
Recent releases
Latest entries from the project's GitHub Releases. Notes are quoted from upstream as written, not rewritten or tested by FollowAgents.
-
@mastra/[email protected] November 4, 2026
Release notes excerpt
Highlights Knowledge v2: page-tree model, richer tools, and server-side sync APIs Knowledge is now a first-class page tree (pages/notes/pins) with major tool upgrades: knowledge_list, line/section/grep-style reads in knowledge_read, list pinning (kind: 'list'), and a link index for counting/listing [[links]]. Apps can also sync external content via server APIs like knowledge.upsertPage()/removePage() and note upserts, with optional vector/hybrid search and chunked embeddings for very long pages. …View on GitHub ↗ -
@mastra/[email protected] November 2, 2026
Release notes excerpt
Highlights MCP Tasks (Storage-backed, restart-safe long-running execution) @mastra/mcp adds experimental Mastra-storage-backed MCP tasks so long-running tools/agents/workflows can return immediately with a task, be polled/answered/cancelled across instances, and survive restarts with shared storage and unified tracing. Background Tasks for Non-Agent Owners (Retention, Concurrency, Queued Resume) …View on GitHub ↗ -
@mastra/[email protected] October 28, 2026
Release notes excerpt
Highlights End-user (per-resource) connections in @mastra/connect tools({ resourceId }) and new userConnectRoutes() enable user-scoped OAuth connections so agents can safely use each end user’s accounts (in addition to global connections), with built-in caching and hard enforcement that one user’s connection can’t serve another user. In-chat “request connection” prompts for missing integrations …View on GitHub ↗
FAQ
Is Mastra tied to one model vendor?
When should I use an agent instead of a workflow?
.then(), .branch(), and .parallel().Can a workflow wait for a human decision?
What licensing issue should adopters check?
ee/ directory is under the Mastra Enterprise License and requires a valid enterprise license for production use.