Case study · Self — Founder
Nmbly
The orchestration layer for AI development workflows.
Year · 2024 — Present
Role · Founder · Designer · Builder
Platform · Tauri (Rust) — local desktop application
Status · Testing with early users. Actively onboarding more for validation.
Product · Nmbly
Built with AI, for the teams building with AI.
The orchestration layer for AI development workflows.
Connect your Jira and Linear tickets to Cursor and other AI agents. Get real-time visibility, MCP-native integration, and workflow control built for product people.
When AI coding agents start doing the work, product managers lose the thread. Tickets sit in Jira. Code gets written in Cursor. The two worlds don't talk. Nobody knows what the agent is actually doing, what's been completed, or what needs to happen next.
Nmbly is the operations layer that fixes that. Not another ticket tracker — an orchestration system that pulls context from Slack, Jira, and Linear, figures out what needs to happen next, moves work forward step by step (investigate → fix → test → deploy), creates follow-up work automatically, and keeps everything connected so nothing gets lost.
I designed it. I built it — using Claude, Claude Code, Cursor, and Figma. A Tauri application that runs locally, MCP-native from the ground up. The product orchestrates the same tools used to create it. I've been working on it for 6-7 months, testing with early users, and building something I genuinely believe product and engineering teams need right now.
Key features
Real-time visibility
See exactly what your AI agents are working on, linked directly to your Jira and Linear tickets
MCP-native
Built on Model Context Protocol for seamless integration with Cursor, Windsurf, and other AI tools
Bi-directional sync
Changes flow both ways — from tickets to code and back, keeping everyone aligned
Product-focused control
Maintain oversight without blocking progress — perfect for non-technical stakeholders
Instant setup
Connect in minutes with one-click integrations for Jira, Linear, and GitHub
Progress tracking
Automated status updates and completion tracking across your entire workflow
AI agents do the work. But nobody knows what they're doing.
AI-assisted development has fundamentally changed how software gets built. Cursor, Claude Code, Windsurf — these tools are genuinely capable of doing the work. The problem is organizational: when an AI agent starts building, the product team loses visibility. Tickets don't update. Context doesn't flow. The PM has no idea if the agent is working on the right thing, what it's done, or what comes next.
Most tools are built to track work. Nmbly is built to move work forward. The difference matters enormously when AI agents are in the loop — because agents need context to work well, and teams need visibility to maintain oversight without blocking progress.
There's a second problem that's specific to design and front-end: design systems become invisible to AI agents. Cursor and Claude Code can read your code, but they can't browse your Storybook, inspect component variants, or understand your design language. The design system sits outside the AI's awareness.
Nmbly solves both.
Three layers. One loop.
Nmbly is built around three interconnected systems — the MCP server that integrates with AI tools, the workflow engine that moves work forward, and the Storybook integration that makes design systems AI-readable. Each is useful alone. Together they create a complete loop from ticket to validated, merged code.
Thread 1 of 3
MCP Server — 31 tools across the stack
Real-time context between tickets and code.
The MCP (Model Context Protocol) server is the heart of Nmbly. It gives AI tools like Claude Code and Cursor direct access to your tickets, workflow state, and project context — so agents work with full awareness of what they're supposed to be building and why.
Currently 31 tools across Nmbly, Claude Code, Cursor, Jira, and Linear. Ticket status updates automatically when code changes. Context flows both ways. The agent knows what the ticket says. The ticket knows what the agent did.
MCP running across Nmbly, Claude Code, Jira, and Linear — 31 tools
MCP in action
Nmbly's MCP running with Claude Code — 31 tools available across the full stack. Agents get ticket context, update status in real time, and stay connected to the planning layer throughout the workflow.
Thread 2 of 3
Workflow engine — investigate → fix → test → deploy
An operations layer, not just a tracker.
Instead of just storing a ticket, Nmbly actually moves work forward. Pull in context automatically from Slack, Jira, and other sources. Figure out what needs to happen next. Move through the full cycle: investigate, fix, test, deploy. Create follow-up work automatically. Keep everything connected so nothing gets lost.
When a task completes, Nmbly automatically creates a PR on GitHub. GitHub Actions runs tests. Status flows back into the Nmbly timeline and the originating ticket. The loop closes without manual intervention.
GitHub workflow automation — PR created, CI triggered, results surfaced
Automatic PR creation
New GitHub workflow automation — when a task completes, Nmbly automatically creates a pull request, triggers CI, and surfaces the results back in the workflow timeline.
Thread 3 of 3
Storybook + MCP — design systems that AI can read
Turn your component library into something queryable.
This is the piece that makes Nmbly distinctive for design-forward teams. Using Claude with the Storybook MCP, AI agents can list components, inspect variants, and understand usage patterns without manually clicking through stories. Your design system becomes queryable — not just readable, but navigable and actionable.
The full loop: create a ticket, start the Nmbly workflow, let the agent implement the change with full awareness of your component library. A PR is created automatically. GitHub Actions runs the Storybook build and tests, and results appear both in the PR and inside the Nmbly timeline. Stories pass. Tests pass. Then merge.
Discovery, implementation, and validation in a single loop. No more guessing if a component change breaks stories or accessibility.
Ticket → agent → accessibility fix → PR → CI → Storybook validation → merge
End-to-end UI shipping
Claude browsing components via the Storybook MCP, a Nmbly workflow kicked off from a ticket, the agent implementing an accessibility fix (aria-label), CI validating Storybook before merge. The same flow works for any UI update.
Integrations
MCP Server
Cursor, Windsurf, Claude Code, any MCP-compatible tool
Project management
Jira, Linear
Code
GitHub
Communication
Slack
Design system
Storybook
Where it goes.
The near-term goal is validation — onboarding early customers, understanding what's most valuable, and building something that brings real value to product and engineering teams navigating AI-assisted development. I'm learning as much as I'm shipping.
The longer-term vision: as AI agents become capable of doing more of the work, the gap between planning and execution grows. The tools that bridge that gap — that give product teams visibility and control without slowing agents down — will become essential infrastructure. Nmbly is an early bet on that future.
This site itself is a proof of concept. It was built using Claude Code and Cursor — the same tools Nmbly orchestrates. The AI agent you're talking to right now is part of the same thread.
What building Nmbly has taught me about designing with AI.
Building Nmbly changed how I think about the relationship between design and engineering. When AI tools are in the loop, the traditional handoff model breaks down — not because designers and engineers aren't collaborating, but because the agent is a third actor that neither the designer's Figma nor the engineer's IDE was built to accommodate.
The Storybook MCP integration came from a real frustration: I kept finding that AI agents would implement UI changes without reference to the design system because they couldn't see it. Making the design system queryable wasn't just a feature — it was a fundamental shift in how AI agents could participate in design-aware development.
I've also learned that building with AI is a design process. The prompts are the brief. The context you provide is the research. The iteration between you and the agent is the design critique. The same skills that make someone a good designer — clarity of intent, good judgment about tradeoffs, knowing when something is done — turn out to make someone a good AI collaborator too.