Hello, Shmastra
Released: April 29, 2025
What is this?
I always wanted to build multi-agent systems and workflows by simply describing what I need in a chat. Let the assistant find the right APIs, hook up MCP servers, connect Google Drive and whatever else — all in one click.
Mastra already lets developers do this in TypeScript with Claude Code. But even for engineers, wiring everything together is a grind.
Shmastra lets anyone — engineers or not — vibe-code agents and entire workflows directly in Mastra Studio. No context-switching, no boilerplate, no setup.
What Shmastra actually is
Shmastra is a Mastra template with a chat widget embedded right inside Mastra Studio. You open the widget, describe what you want in plain English, and Shmastra's coding agent writes the TypeScript, runs a dry-run build, and hot-reloads the project — all without you leaving the browser.
npx create-mastra@latest --template https://github.com/just-ai/shmastra
npm run devThat's it. Two commands and you're talking to the widget.
Everything the widget generates is real, editable TypeScript under src/mastra/. Shmastra is not a black box — it's a coding assistant that writes code into your own project. Pop open VS Code at any point and keep going by hand.
Powered by mastracode
The coding agent inside Shmastra is built on top of mastracode — a harness that gives an LLM structured, skill-based access to a Mastra project. Each capability (creating agents, wiring channels, building web apps) is a separate skill file that teaches the model exactly what to generate and how to register it. This is what makes the widget reliable: instead of free-form code generation, mastracode constrains the model to well-defined patterns and validates the output with a dry-run build before hot-reloading.
Build anything by describing it
The widget knows how to build the full Mastra surface:
Agents. Describe the agent's name, persona, and goals. The widget writes the agent file, registers it, and hot-reloads — the agent shows up in Studio in seconds.
Workflows. Describe a multi-step process and the widget composes it as a Mastra workflow with proper step dependencies, input/output shapes, and suspend/resume support.
Inspect & fix. When something behaves wrong, the widget already has context of whatever is open in Mastra Studio — the current agent, its last run, logs, traces. It sees what's happening without you having to paste anything, analyses the root cause, and applies the fix.
Give your agents superpowers
Agents are only as useful as their tools. A few prompts unlock the good stuff:
Working with files. Upload a PDF, DOCX, XLSX, or HTML file and immediately ask questions about it. Shmastra wires up a RAG pipeline under the hood — chunking, embedding, and retrieval included.
Web search. Ask for web search and the agent gets a live-results tool powered by Brave Search. No API wrangling, no middleware.
Headless browser. Give an agent a browser — Playwright-backed, running inside the sandbox — and it can scrape pages, fill forms, and interact with the web like a person would.
MCP servers. Point Shmastra at any MCP server URL and the coding agent will install it and wire it into your agent as a tool provider. Notion, Linear, GitHub — anything MCP-compatible just works.
Composio toolkits. With a Composio API key you unlock 200+ third-party integrations: Gmail, Google Drive, Slack, Notion, HubSpot, and more. One prompt to add them; one-click OAuth to authorize them.
Ship it somewhere
Building an agent in Studio is great. Talking to it from the real world is better.
Channels. Connect any agent to a messaging platform. This initial release ships with 14 channel adapters out of the box:
- Discord, Slack, Microsoft Teams, Mattermost, Google Chat, Webex
- Telegram, WhatsApp, Zalo, Zernio (social DMs)
- Email via Resend
- GitHub (issues and PR comments as triggers)
- Linear (issue and comment events)
- Liveblocks (collaborative document comments)
Each channel takes a handful of env vars and a prompt like "Connect the support agent to my Telegram bot" — the widget handles the rest.
Web apps. Ask the widget to build a frontend and it generates a self-contained HTML/JS/CSS app served directly from the Mastra route layer. Charts, dashboards, forms — anything that can be expressed in vanilla HTML.
Rich chat replies. Agents can respond with formatted HTML, embedded iframes, and interactive widgets — not just plain text — right inside the chat widget.
Shmastra Cloud: share with your team
Running Shmastra locally is great for building. Shmastra Cloud is for when you want everyone at your company to vibe-code agents from a browser — no Node, no CLI, no local setup.
You deploy one Next.js app to Vercel. From that point, every teammate who signs in through WorkOS gets their own E2B sandbox with Mastra Studio and Shmastra pre-installed. Private filesystem, private processes, private environment — but the same widget experience you have locally.
The Cloud adds things you'd otherwise have to build yourself:
Virtual-key gateway. Each sandbox talks to AI providers through a virtual key that Cloud swaps for your real API key at the edge. Users never see your keys; you control all provider access from one place.
Manage UI. An admin dashboard you run from your laptop. It lists every sandbox, lets you stream logs, browse files, open a terminal, and chat with a Claude session running inside any sandbox. Roll out updates to all sandboxes in one click.
Healer agent. A second process that runs alongside each Shmastra instance. When the Mastra dev server crashes or gets stuck, the healer diagnoses the problem and repairs it automatically — no pager, no intervention.
Workflow scheduling. Any workflow can run on a cron. Define the schedule in the widget, results and full traces are stored per run, and the Stats and Trace tabs in the Manage UI give you a timeline of every execution.
Patch system. Ship migrations to all sandboxes without rebuilding the E2B template. Patches are tiny shell scripts that run once and mark themselves complete — ideal for config changes, dependency updates, or one-off data migrations.
What's next
This is the initial release. The foundation is solid but there's a lot more coming.