One Next Step

• 

Inspiration

Having something you want to do does not always mean having the capacity to begin. A task can feel overwhelming, uninteresting, or difficult to approach when energy and focus are limited.

One Next Step grew from wanting a gentler way to start. It helps adults who experience executive-function challenges identify one manageable action that fits the capacity they have right now.

What it does

ONS asks users what they want to work on, then checks their energy, focus, emotional room, available time, and current context. It preserves their chosen task and offers support based on what makes starting difficult.

The demonstration follows a simple example: “I want to write a novel.” The user has low energy, limited focus, ten minutes, and is currently at school. Choosing School keeps the novel as the selected task.

When the user selects “It feels boring,” ONS suggests writing one outrageous line of dialogue a character might say. The exercise takes two minutes and does not have to become part of the novel.

The user leaves with a concrete place to begin, without streaks or penalties for stopping.

How we built it

ONS uses a React and TypeScript interface, a Cloudflare Workers runtime, and D1 persistence.

For the Alexa+ track, I implemented a working MCP server using the MCP SDK and Streamable HTTP. Six tools support brain-dump capture, capacity checks, recommendations, task unfreezing, task saving, and step outcomes.

The signed-in browser demonstration calls that server through an authenticated session. An MCP activity panel shows initialization, tool discovery, and actual tool calls without exposing private request contents.

The demonstrated orchestration and first-step prompts use deterministic application rules and do not call a language model. Live Alexa+ account linking and Echo-device interaction have not been validated.

Pre-existing project and hackathon work

The ONS concept and earlier app work existed before the hackathon window. During the window, I significantly developed the project through the MCP integration, authenticated browser demonstration, session protections, task-preservation improvements, blocker-specific creative-writing prompts, and regression testing.

The video demonstrates these additions through the complete task-to-first-action journey and the MCP activity panel.

Challenges we ran into

One challenge was preserving user intent throughout the workflow. An early version could recommend an unrelated saved task after the user chose a context. I changed the flow to save the entered task and carry its identifier through capacity checking and recommendation.

Another challenge was making the resulting action useful. A generic instruction to work on the easiest piece was too vague. Creative-writing tasks now receive concrete prompts that respond to the selected blocker.

I also separated browser-session authentication from the external OAuth bearer endpoint and added checks for task ownership, request origin, session expiry, and retry-safe writes.

Accomplishments that we're proud of

ONS connects a working MCP server to a browser experience that preserves the user’s chosen task and returns a manageable starting action.

The focused Version 10 regression suite passed 22 tests, and I manually verified the hosted novel, School, and boredom journey.

What we learned

I learned that preserving user intent is as important as making a successful tool call. Context should inform the support while keeping the chosen task in focus.

Testing also showed that a smaller instruction must be concrete and respond to what makes starting difficult. Automated tests and real browser walkthroughs helped verify both the technical behavior and whether the result made sense.

What's next for One Next Step

I plan to expand support for more task types and gather feedback from adults with executive-function challenges. That feedback will guide improvements to the starting actions, accessibility, and overall experience.

I also want to validate external Alexa+ account linking and voice interaction, building on the working MCP server demonstrated here.

The longer-term goal is to offer ONS through both a website and a mobile app, helping people find one manageable action that fits their capacity.

Built With

• cloudflare
• cloudflared1
• css
• mcp
• react
• tailwind
• typescript
• vite
• workers
• zod