Zyk – Claude-Native Workflow Automation

TL;DR

TL;DR: Zyk flips the workflow automation model—instead of dragging nodes in an editor, you talk to Claude and it builds, deploys, and runs durable workflows on Hatchet under the hood.

Source and Accuracy Notes

⚠️ This section is MANDATORY. All links must be verified from actual source, not guessed.

• Project page: zyk.dev ← MUST visit and verify
• Source repository: github.com/zyk-hq-dev/zyk ← MUST read README
• License: MIT (verified via GitHub API license.spdx_id)
• HN launch thread: news.ycombinator.com/item?id=47586452
• Source last checked: 2026-06-23 (commit main, 6 stargazers)

What Is Zyk?

Most workflow automation tools add AI as an afterthought—bolting a chat box onto an existing node editor. Zyk takes the opposite approach: Claude is the interface. You describe what you want in plain language, Claude generates real TypeScript code, and Zyk deploys it on Hatchet with durable execution, retries, scheduling, and human-in-the-loop approval all built in.

The key differentiators from the README:

• No node editor. No connectors to configure. If a connector does not exist, you’re not blocked—Claude writes the HTTP calls directly.
• Real TypeScript, not a DSL. Workflows are plain TypeScript files you own. If you outgrow Zyk, run them on Hatchet directly without changing a line.
• Durable execution. Workflows can pause mid-execution and wait days for a human signal, then resume exactly where they left off. No polling loops, no external state store.
• Deploy on Railway in under 10 minutes. No Docker required locally. One docker compose up if you want to run locally.

Setup Workflow

Step 1 — Deploy on Railway

Use the one-click Railway deploy. Railway provisions PostgreSQL, Hatchet Engine, and the Zyk MCP Server automatically:

→ Deploy on Railway

After deploy, copy two values from the Railway dashboard:

1. ZYK_API_KEY — from the zyk service Variables tab
2. Zyk MCP Server URL — from Settings → Networking → Public URL (looks like https://<zyk>.up.railway.app)

Step 2 — Connect Claude

Claude Desktop (recommended for alpha):

Open Settings → Developer → Local MCP Servers → Edit Config and add the zyk entry inside "mcpServers":

{
  "mcpServers": {
    "zyk": {
      "command": "npx",
      "args": ["-y", "zyk-mcp", "--proxy", "https://<zyk>.up.railway.app/mcp"],
      "env": {
        "ZYK_API_KEY": "your-secret-key"
      }
    }
  }
}

Fully quit and restart Claude after saving. Node.js must be installed.

Claude in browser: go to Settings → Connectors → Add custom connector, paste the Railway URL and API key. No local setup needed.

Claude Code: add to .mcp.json:

{
  "mcpServers": {
    "zyk": {
      "type": "http",
      "url": "https://<zyk>.up.railway.app/mcp",
      "headers": {
        "Authorization": "Bearer your-secret-key"
      }
    }
  }
}

Step 3 — Verify It Works

Open the Zyk dashboard (https://<zyk>.up.railway.app) and Hatchet UI (https://<hatchet-lite>.up.railway.app). Default Hatchet login: [email protected] / Admin123!! — change this password after first login.

In Claude, ask: “List my workflows”

Expected response:

No workflows registered yet. Use create_workflow to create your first workflow.

Then try the full loop:

1. “Create a workflow that asks me what my favourite colour is and logs my answer”
2. “Run it”
3. Open the Zyk dashboard Tasks tab — a question appears within seconds
4. Answer it — the run completes and the result appears in the Hatchet UI

How It Works

You (natural language)
    |
Claude
    |  MCP over HTTP
Zyk MCP Server  (Railway)
    |-- /app/workflows/registry.json   workflow registry
    +-- worker subprocess per workflow
            |  gRPC
        Hatchet Engine  (Railway - :8080 UI, :7077 gRPC)
            |
        PostgreSQL  (Railway - run history, scheduling, durable state)

Scheduling: cron expressions live inside the workflow TypeScript code (on: { cron: "0 8 * * *" }). Hatchet owns scheduling entirely.

Human-in-the-loop: workflows use workflow.durableTask() + ctx.waitForEvent() to pause durably in Hatchet’s database. Respond via the Zyk dashboard or Slack, and the step resumes exactly where it left off.

Slack integrations: set the Slack app’s Interactivity Request URL to https://<zyk>.up.railway.app/slack/interactions.

Practical Evaluation Checklist

• [ ] One-click Railway deploy succeeds without errors
• [ ] ZYK_API_KEY and MCP Server URL are correctly copied from Railway
• [ ] Claude Desktop MCP config saves without crashing
• [ ] “List my workflows” returns the expected empty-state message
• [ ] Creating and running a simple workflow completes without error
• [ ] Human-in-the-loop pause and resume works correctly
• [ ] Hatchet UI shows the workflow run history and diagram
• [ ] Default Hatchet password is changed after first login

Security Notes

• Hatchet default credentials ([email protected] / Admin123!!) are public knowledge. Change the password immediately after first login.
• ZYK_API_KEY grants full access to your workflows—treat it like a database password, not an environment variable.
• The Zyk dashboard is public-facing on the Railway public URL; ensure access controls are appropriate for your use case.
• Workflows run as TypeScript subprocesses—code generation by Claude means inputs to those workflows should be treated as untrusted.

FAQ

Q: Is this production-ready? A: The README explicitly states alpha is live but not production-ready. Treat it as early-stage software.

Q: Does it work without Claude? A: No. Zyk is designed around Claude as the primary interface. It connects via the MCP protocol.

Q: Can I self-host without Railway? A: Yes. See CONTRIBUTING.md in the repo for a local Docker Compose setup. This requires Docker on your machine and a public HTTPS URL for Slack webhooks to work without ngrok.

Q: How does it compare to Temporal or Inngest? A: Zyk builds on Hatchet (a Temporal-like durable execution engine) rather than self-hosting Temporal directly. Hatchet was chosen for its single-Docker-image simplicity—no Kafka or Cassandra dependency.

Conclusion

Zyk is an interesting flip on the AI workflow tool category: instead of making an existing automation tool slightly smarter, it makes Claude the builder and layers durable execution underneath. The trade-off is tight coupling to Claude, but for teams already living in Claude Desktop, this could replace a separate automation platform entirely. Self-hosting on Railway with zero Docker locally is a genuine ergonomic win.

The alpha status is the main caveat—watch the repo for stability improvements before trusting it with anything important.