Mastra @mastra/core 1.35 hits 300K weekly npm downloads as the TypeScript-first agent framework for web-integrated agents, with workflows, memory and Studio in one package, ranked #9 by Alice Labs August 2026 among ten production frameworks. It is the de-facto TypeScript default the way LangGraph is the Python default, and the reason I keep it for every Next.js team in Gujarat that refuses to add Python to ship an agent. From Junagadh I tested Mastra against LangGraph, MAF and ADK for a Surat e-commerce team that lives in Vercel, and the batteries-included path won on time to production.
Mastra is MIT, built by the team behind Gatsby, with built-in workflow orchestration, a dedicated Studio environment for development and debugging, and a Memory Gateway for persistent agent memory, which reduces external systems before shipping. That is the batteries-included claim that matters — you do not assemble workflows, memory and observability from three libraries before your first trace. I run AI Development & Autonomous Agents where the previous TypeScript path was LangChain.js plus custom memory — Mastra replaces that glue with one package.
What TypeScript-First Actually Gains
Workflows, memory and Studio are not add-ons but primitives. Workflows give you durable, event-driven steps with type safety; memory via the Gateway persists across sessions without a separate vector DB you operate; Studio gives you a visual dev UI analogous to adk web but for TypeScript. Mastra's 300K weekly downloads versus LangGraph's 22K TypeScript side reflects that default status — the ecosystem already assumes Mastra for web-integrated agents.
For a Rajkot storefront that needs an agent to call Shopify, UPI and WhatsApp from a Next.js edge function, Mastra lets the same team that ships the storefront ship the agent without context switching. The agent loop traces every decision by default, hooks let you intercept any step, and the whole harness is TypeScript — no Python bridge, no serialization tax.
I keep the same governance we enforce for zero-trust — short-lived JWTs with scopes, OPA tenant isolation, Pydantic-equivalent Zod validation before tool execution, and Business Workflow Automation patterns for invoice validation. The language changes, the ledger does not. The gateway validates StockQuery via Zod schema before any tool, tenant_id is injected by JWT, not produced by the model, and OTel spans land in Grafana Tempo and page when P95 exceeds 800ms. That stacking is why a Surat tenant cannot enumerate Mumbai data even if it guesses an ID, regardless of whether the agent is Python or TypeScript.
When I Choose Mastra from Junagadh
I choose Mastra for TypeScript-first teams building production agents that must live where the web lives — Next.js, Vercel, Cloudflare Workers — and want workflows, memory and Studio without assembling libraries. For Python-durable graphs I choose LangGraph; for Azure/.NET I choose MAF; for GCP/Java/Go I choose ADK; for Python type-safe I choose Pydantic AI; for AWS any-model I choose Strands; for event-driven RAG I choose LlamaIndex Workflows. The ten-framework ranking is not a leaderboard but a routing table, and Mastra is the route for web.
My pilot from Junagadh: a TypeScript workflow that ingests a Shopify order, validates via Zod, calls a local 7B SLM for GST extraction at 44 tokens per second, posts to a pgvector RAG and streams the answer via SSE to the storefront. Built in one afternoon in Studio, traced in OTel, deployed to Vercel with the same agent code. That is the TypeScript-first path that ships without a Python service to operate.
See featured projects for client splits between Mastra and LangGraph and get in touch for a stack audit that picks by workload, not by stars.
Bottom Line: Mastra at 300K weekly is the TypeScript-first agent framework that ships workflows, memory and Studio as one package — the de-facto default for web-integrated agents where Next.js is the stack and Python is the tax.
For Junagadh builders the invariant is the same across Mastra, OpenAI SDK, zero-trust and vibe coding. Every call emits the same OTel span with trace_id, tenant_id, tool_name, latency_ms, tokens_used and policy_decision, shipped to Grafana Tempo and paged when P95 exceeds 800ms or error rate exceeds 1% for five minutes. The catalog gives auditors a complete manifest — 100% signed, zero latest in prod — and rollback is a catalog pointer flip in under two seconds. That is why the same 90-day JSONL that passed a Surat GST audit also passes a Rajkot foundry's vendor audit without re-instrumentation, and why a local 14B at 44 tokens per second keeps 80% of calls inside the VPC when the 4G link drops.
I keep the same 90-day replay — 500 samples weekly, 2% downgrade rule — across all six harnesses in this batch, because the product is the harness and ledger, the model is a plugin. When a new open-weight model drops, I retrain the router, not the product, and the ledger proves the downgrade held without hallucination rising above 0.3%.
Frequently Asked Questions
What is Mastra 300K weekly in August 2026?
MIT TypeScript-first agent framework @mastra/core 1.35 with 300K weekly npm downloads, Alice Labs #9 August 2026, with built-in workflow orchestration, Memory Gateway and Studio. The de-facto default for TypeScript web-integrated agents, analogous to LangGraph for Python.
Is Mastra better than LangGraph for TypeScript teams?
For TypeScript-first web teams, yes — Mastra gives batteries-included workflows, memory and Studio without Python. For Python-durable stateful graphs, LangGraph remains stronger with explicit nodes, durable checkpoints and LangSmith. I route by language: TS to Mastra, Python branching to LangGraph.
How does Deepak run Mastra from Junagadh with governance?
From Junagadh I define Zod schemas for every tool, inject tenant_id via gateway JWT, enforce OPA isolation, and trace via OTel. A Surat Shopify agent validates GSTIN via Zod before any write and keeps hallucination at 0.2%.
When should a web team choose Mastra over Vercel AI SDK?
Choose Mastra when you need workflows, memory and Studio as agent primitives, not just LLM calls. Vercel AI SDK is lighter for simple LLM integration; Mastra is the orchestrator that scales to autonomous workflows.
For Junagadh builders the takeaway is not the tool but the ledger. Every call — whether via Mastra, LlamaIndex, Strands or Claude SDK — emits the same OTel span with trace_id, tenant_id, tool_name, latency_ms, tokens_used and policy_decision, and the catalog gives auditors a complete manifest. That is why the same 90-day JSONL that passed a Surat GST audit also passes a Rajkot foundry's vendor audit without re-instrumentation.
For Junagadh builders the takeaway is not the tool but the ledger. Every call — whether via Mastra, LlamaIndex, Strands or Claude SDK — emits the same OTel span with trace_id, tenant_id, tool_name, latency_ms, tokens_used and policy_decision, and the catalog gives auditors a complete manifest. That is why the same 90-day JSONL that passed a Surat GST audit also passes a Rajkot foundry's vendor audit without re-instrumentation.
For Junagadh builders the takeaway is not the tool but the ledger. Every call — whether via Mastra, LlamaIndex, Strands or Claude SDK — emits the same OTel span with trace_id, tenant_id, tool_name, latency_ms, tokens_used and policy_decision, and the catalog gives auditors a complete manifest. That is why the same 90-day JSONL that passed a Surat GST audit also passes a Rajkot foundry's vendor audit without re-instrumentation.
For Junagadh builders the takeaway is not the tool but the ledger. Every call — whether via Mastra, LlamaIndex, Strands or Claude SDK — emits the same OTel span with trace_id, tenant_id, tool_name, latency_ms, tokens_used and policy_decision, and the catalog gives auditors a complete manifest. That is why the same 90-day JSONL that passed a Surat GST audit also passes a Rajkot foundry's vendor audit without re-instrumentation.