shiftweave

A shift change shouldn’t mean starting from zero. ShiftWeave, an Alexa+ MCP server, designed for nurses, turns “Where did we leave off?” into “I’ve got this.”

Comment

Inspiration

A shift ends. Unfinished work still needs an owner.

We built ShiftWeave to help the next nurse understand what remains, what needs verification and what they are accepting responsibility for—without reconstructing the previous shift.

What it does

ShiftWeave combines a conversational nursing interface with an MCP server that makes its workflow capabilities accessible to compatible assistants.

Nurses can retrieve patient and task information, review conflicting records, confirm supported updates and accept handoffs. Supported interrupted workflows resume from saved progress.

The aim is simple: clear context, explicit ownership and continuity across shifts.

How we built it

We combined React, a TypeScript backend, AWS Bedrock for language understanding and MCP tools over Streamable HTTP.

SQLite preserves assignments, handoffs, audit records and recovery checkpoints. Backend services validate requests and ground workflow answers in retrieved information.

Epic FHIR integration remains available alongside a separate synthetic hospital preview. External MCP clients default to the Epic path; preview interactions use explicitly scoped simulation data.

Challenges we ran into

Natural conversation had to work alongside precise execution: understanding intent without inventing facts, transferring responsibility without assuming consent, and recovering interrupted work without repeating changes.

Our tested Epic sandbox access did not cover several nursing workflow scenarios. We preserved that integration and added clearly labeled synthetic workflows backed by actual database operations.

Accomplishments we're proud of

We connected conversation to working task retrieval, discrepancy review, confirmed updates, handoff acceptance and checkpoint recovery.

Local checks exercised database persistence, retries, recovery and source isolation. Live Bedrock-to-MCP replays demonstrated contextual conversations.

The voice interface makes these workflows approachable; the MCP server makes the capabilities reusable beyond that interface.

What we learned

A natural answer matters only when the system can support it with evidence.

Language understanding, usable interfaces and reliable backend execution serve different purposes. Trust depends on bringing them together—with clear permissions, confirmation, durable state and honest uncertainty.

Simulation helped us test realistic scenarios while keeping their limitations visible.

What's next for ShiftWeave

Next, we will broaden compatible MCP client testing, validate deployed voice interactions, measure latency and strengthen preview controls and recovery procedures.

Institution-approved workflow mappings, successful Epic publication and clinical deployment remain separate validation milestones.

Built With

Share this project:

Updates

Submission history