Mastra
Also known as: Mastra Studio, Mastra Platform
Open source TypeScript framework for agents and workflows, with the Mastra Platform for hosted Studio, observability, evals and deployment.
Mastra is an open source TypeScript framework for building AI agents and applications, released under the Apache 2.0 license and installed from npm. An agent is defined in one place with its instructions, model, tools and memory, and the model is set as a provider and model string in Mastra's model router, which reaches thousands of models. Tools are typed functions the agent can call, and the documentation also covers MCP connections, subagents, skills, sandboxes and channel integrations.
Workflows compose typed steps, retries and branches into durable multi-step processes, and a harness coordinates multi-mode agents over shared state, threads and storage; Mastra says the same harness runs Mastra Code, its coding agent. Memory gives agents durable context through message history, observational memory and semantic recall. Agents can also be given a browser, through a local Playwright-based provider or hosted browser services, with a live screencast of the session in Mastra Studio. Mastra also ships Factory, a system of specialized agents that takes software from issue to production.
The Mastra Platform hosts Studio, deployments, observability and evals. Traces record each agent run step by step alongside logs and metrics, and scorers, datasets and experiments score agent behavior across runs. The Platform has a free Starter tier, a Teams tier that adds SSO, multiple teams and SOC 2 documentation, and an Enterprise tier with RBAC, audit logs, support SLAs and on-premises or own-cloud deployment. A separately licensed self-hosted Enterprise edition adds RBAC, SSO, IAM and network policy integration and keeps traces, prompts and outputs inside the customer's VPC.
Vendor details
Canonical URL
https://mastra.ai/
Category
Agent infrastructure
Company status
independent
Use cases & customers
Target customers
Deployment options
In practice
Your team works in TypeScript but every serious agent framework assumes Python. Mastra is TypeScript-native, giving you agents, tools, memory, and workflows in one stack that integrates cleanly with Next.js and Node.
Part of your process needs an agent's judgment and part needs deterministic steps in a fixed order. Mastra's workflows mix agent reasoning with plain code in type-safe control flow, and can pause for human approval before resuming.
You don't want to commit to one model provider or wire up evals and tracing yourself. Mastra routes across thousands of models through one interface and ships built-in scorers, guardrails, and observability to make agents production-ready.
Sources & related URLs
Related / legacy domains
Agentic Index coverage score
14.0 / 14 capabilities · 100%
| Integrations & Tool Calling | Full |
|---|---|
|
Custom tools are defined with createTool, a typed input schema and an execute function that can call any system, and are passed to the agent. Mastra also connects to MCP servers and offers a large integrations catalog. SourceMastra, mastra.ai/docs and mastra.ai/llms.txtread 2026-09-21 |
|
| Workflow Orchestration | Full |
|
Workflows compose typed steps, retries and branches into durable multistep processes, with agents and tools callable as workflow steps, so fixed steps and agent steps run in one process. A harness coordinates multimode agents over shared state. SourceMastra, mastra.ai homepage and mastra.ai/pricingread 2026-09-21 |
|
| Knowledge Grounding & RAG | Full |
|
Workspace search keeps searchable content in a separate index, queried by BM25 keyword, vector or hybrid search, and content from files, databases, APIs or other application sources can be added to it. Search and indexing tools are given to the agent, so it can look up the customer's own content as it works. SourceMastra, mastra.ai/docs/sandbox/searchread 2026-09-21 |
|
| Human Oversight & Guardrails | Full |
|
A tool marked for approval pauses before it executes and a person approves or declines it, with an optional reason returned to the model; approval can be required for one tool, for every call in a request, or decided per call by a function over the tool name and arguments, and a running tool can also suspend for confirmation, so low and high risk actions can run under different autonomy modes. SourceMastra, mastra.ai/docs/agents/human-in-the-loopread 2026-09-21 |
|
| Security, Identity & Governance | Full |
|
SSO and multiple teams come with the Teams plan, RBAC and audit logs with Enterprise, and RBAC, SSO, IAM and network policy integration with the licensed self hosted edition. SOC 2 documentation is offered on the Teams plan, and a hosted trust center is published. SourceMastra, mastra.ai/pricing and trust.mastra.airead 2026-09-21 |
|
| Observability & Auditability | Full |
|
Traces record each agent run step by step, including input processors, model calls, steps and tool calls, alongside logs and metrics. On the Platform, traces are kept 15 days on Starter and six months on Teams, Enterprise adds audit logs, and OpenTelemetry and Datadog exporters are available. SourceMastra, mastra.ai homepage and mastra.ai/pricingread 2026-09-21 |
|
| Memory & State Persistence | Full |
|
Memory gives agents durable context across sessions through message history, observational memory and semantic recall over thread aware storage. On the Platform, stale threads are kept 15 days on Starter and six months on Teams. SourceMastra, mastra.ai homepage and mastra.ai/pricingread 2026-09-21 |
|
| Deployment & Data Residency | Full |
|
The Apache 2.0 framework builds and hosts agents anywhere, and the licensed self hosted Enterprise edition keeps traces, prompts and outputs inside the customer's VPC. Enterprise also offers the Mastra Platform deployed on premises or in the customer's own cloud. SourceMastra, mastra.ai/pricingread 2026-09-21 |
|
| Prebuilt Agents, Templates & Packs | Full |
|
Mastra ships Factory, a system of specialized agents that takes software from issue to production, and Mastra Code, a coding agent running on the Mastra harness. Each is a product in its own right. SourceMastra, mastra.ai homepageread 2026-09-21 |
|
| Triggers & Channel Coverage | Full |
|
A schedule runs an agent on a cron cadence, persisted across restarts and redeploys, and on each fire Mastra sends the prompt into a thread as a signal or as a standalone run, so the agent wakes without a person starting it. The same service schedules workflows, and channel integrations cover Slack, GitHub, Microsoft Teams and WhatsApp. SourceMastra, mastra.ai/docs/harness/schedules and mastra.ai/llms.txtread 2026-09-21 |
|
| Model Flexibility & Routing | Full |
|
The developer sets each agent's model as a provider and model string in Mastra's model router, which reaches thousands of models across providers such as OpenAI, Anthropic and Google, and the Platform gateway also accepts the customer's own keys. SourceMastra, mastra.ai/docs, homepage and mastra.ai/pricingread 2026-09-21 |
|
| APIs, SDKs & MCP Extensibility | Full |
|
Mastra is a TypeScript SDK installed from npm as @mastra/core, and agents and workflows registered on a Mastra server are exposed to outside callers through a Mastra Client, custom API routes and A2A, with a full API reference. SourceMastra, mastra.ai/docs, homepage and mastra.ai/llms.txtread 2026-09-21 |
|
| Testing, Debugging & Optimization | Full |
|
Scorers, datasets and experiments score the customer's agent on named checks such as correct tool selection and correct answer, with results compared across runs. Mastra offers built in and custom scorers, gates and verdicts, and evals that run in CI. SourceMastra, mastra.ai homepage and mastra.ai/llms.txtread 2026-09-21 |
|
| Browser & Computer Use | Full |
|
Assigning a browser to an agent gives it tools to open URLs, select and interact with page elements, fill forms and read content, through the Playwright-based AgentBrowser provider that launches a local browser, and Mastra Studio shows a live screencast of the agent's session. SourceMastra, mastra.ai/docs/browserread 2026-09-21 |
|
The Agentic Index coverage score grades every vendor Full, Partial or Not documented against the same 14 buyer facing capabilities, from public evidence only. Each capability links to how all vendors in the index score on it. How this evidence is graded
Recent platform changes
Any environment on Mastra's hosted platform can now get a Postgres database that accepts connections only from inside the same private network. It is created with one CLI command or at deploy time, placed in the environment's region and wired in automatically.
Bears on: Memory / state
View sourceMastra released @mastra/mcp 2.0.0, whose servers support only the July 28, 2026 MCP revision and require @mastra/core 1.68 or newer. Tools that need caller input now suspend and resume through input_required continuations with signed request state. The release removes legacy initialization, session headers, standalone HTTP plus SSE transport, and earlier elicitation APIs.
Bears on: Human approval / guardrails
View sourceMastra introduced Filesystem Skills for Workspaces, enabling reusable skills for multi-agent systems.
Bears on: Agent capability
View sourcePricing
From $250/mo · free tier
hybrid
Included quota
Framework: free under Apache 2.0, self-hosted. Starter ($0): 100K observability events, 24 CPU hours, 15-day retention, 10GB egress, unlimited users, deployments and projects. Teams ($250/month): 1M events, 250 CPU hours, six-month retention, 100GB egress, multiple teams, SSO and SOC 2 documentation. Enterprise: custom volume and retention, RBAC, audit logs, support SLAs.
What is public
The framework is free under Apache 2.0. The Mastra Platform publishes Starter at $0 and Teams at $250 per month, each with stated allowances and overage rates for observability events, CPU time, egress, memory tokens and retrieval storage; Enterprise is custom. A licensed self-hosted Enterprise edition is priced by sales. Model usage is billed at market rate plus 5.5% through the gateway, or directly by the provider on your own key.
Billing mechanics
Two paths: self-host the Apache 2.0 framework at no charge, paying your own infrastructure and model costs, or use the Mastra Platform, with a monthly tier fee plus metered usage on observability events, CPU time, egress, memory tokens and retrieval storage. A licensed self-hosted Enterprise edition is sold for one flat annual fee with no per-trace or per-seat metering.
Cost watchouts
Platform usage is metered beyond each tier's allowance on observability events, CPU time, data egress, memory tokens and retrieval storage; a 24/7 persistent server is $100 per project; model tokens through the Mastra gateway carry a 5.5% markup over market rate unless you bring your own key.
Variable cost rationale
A flat tier fee sits under usage metering on several dimensions, and model token spend scales with agent runs on top.
Additional watchouts
Metering spans several dimensions, so production cost tracks trace volume and CPU time; SSO starts at Teams, while RBAC and audit logs start at Enterprise.
Overage / add-ons
Starter: $10 per 100K observability events, $0.35 per CPU hour, $0.10/GB egress. Teams: $8 per 100K events, $0.25 per CPU hour, $0.10/GB egress. Memory tokens $10 per 1M beyond the included amount; retrieval storage $20/GB; persistent 24/7 server $100 per project; gateway tokens at market rate plus 5.5%, or bring your own key.
Sales call required
Mixed (some tiers require a call)
Free / trial
Free tier
Lowest paid plan
Teams $250/mo
Commercial notes
Mastra sells two things beside the free framework: the hosted Mastra Platform (Starter, Teams, Enterprise) and a licensed self-hosted Enterprise edition for teams that need RBAC, SSO, IAM and network policy integration with data kept in their own VPC.
Key ambiguities
Enterprise pricing for the Platform and for the licensed self-hosted edition is by sales; Starter and Teams publish their allowances and overage rates.
Cancellation / refund
The framework has no contract; Platform Starter is free and Teams is self-serve at $250 per month; Enterprise and the licensed self-hosted edition are contracted through sales.
Support SLA / resale
Enterprise support includes architecture and design meetings, a shared Slack channel, separate service SLAs with weekday business-hours coverage, on-prem platform deployments and SRE support.
Missing data
Enterprise pricing (Platform and licensed self-hosted) is not published.
Related vendors
- AgentOps — Agent observability and debugging platform: open source SDKs trace…
- Agno — Python agent framework and AgentOS runtime (formerly Phidata) for…
- AIsa — Resource and payment gateway for AI agents: one key to 110+ models…
- AlphaBitCore — AI control plane for regulated financial firms: one gateway enforces…
- Anchor Browser — Cloud hosted browser infrastructure that lets AI agents operate real…
- Apify — Cloud platform and marketplace of more than 73,000 ready-to-run…
Alternatives to Mastra
The closest documented capability profiles to Mastra among agent infrastructure platforms tracked by Agentic Index, ordered by similarity on the same 14 point evidence the rankings use. No vendor pays for placement.
- Agno13.0 / 14A lighter documented profile than MastraMastra vs Agno →
- Tray.ai13.0 / 14A lighter documented profile than Mastra
- Kestra12.5 / 14A lighter documented profile than Mastra
- Haystack12.0 / 14A lighter documented profile than Mastra
- LlamaIndex12.0 / 14A lighter documented profile than Mastra
- Pydantic AI12.0 / 14A lighter documented profile than Mastra
Similarity is computed from each vendor's Agentic Index coverage score evidence, axis by axis, not from the totals. How this evidence is graded