ProtocolOS

Inspiration

Research protocols define exactly what must happen, yet daily operations are often split across documents, spreadsheets, calendars, and task systems. Coordinators spend valuable time reconciling those sources to answer a deceptively simple question: what does this subject need next? ProtocolOS began with the idea that a protocol should be an operational workflow that people and browser agents can understand together—not just a static document.

What it does

ProtocolOS turns research protocols into shared, agent-ready workflows. It organizes multiple synthetic studies, subjects, stages, visits, assessments, samples, tasks, and completion criteria in one polished workspace.

Coordinators can:

• See today's priorities across multiple studies
• Understand a subject's current stage, visit, and outstanding work
• Build and publish a visual protocol workflow
• Generate a visit checklist from the published workflow
• Complete tasks while keeping related records synchronized
• Check whether a subject is ready to advance, with explicit reasons when advancement is blocked

All demo records are fictional and use generic IDs such as SUB001. ProtocolOS contains no real protected health information.

How WebMCP creates real utility

Without WebMCP, an agent would have to infer operational state from pixels and page text. That is fragile when a missed assessment, sample, or sign-off can change whether a visit is complete.

ProtocolOS registers fourteen imperative tools through document.modelContext.registerTool(). They cover portfolio summaries, priorities, workflow inspection and publishing, subject status, assessments, visits, samples, missing tasks, readiness, checklist creation, task completion, and guarded stage advancement.

Read operations are marked read-only. Mutations are explicit and update the same React state shown to the coordinator. Every response includes structured results, synthetic-data indicators, workflow provenance, and a compact technical receipt.

Human and agent collaboration

A coordinator can ask, “What does SUB001 need today?” The agent retrieves the current stage and visit, incomplete tasks, assessments, sample requirements, and next action. It can create the checklist and complete an approved item. The visible workspace updates immediately, and the activity trail records who changed what and when.

Crucially, the model does not decide whether a subject advances. ProtocolOS evaluates the published completion gates and returns a clear allowed-or-blocked decision with the exact remaining requirements. The workflow remains the source of truth while the agent accelerates coordination.

How we built it

ProtocolOS is a responsive React and TypeScript application styled with Tailwind CSS and hosted with OpenAI Sites. Synthetic seed data powers a multi-study dashboard, subject records, a visual workflow builder, visit checklists, activity history, and an agent workspace. WebMCP actions are registered directly in the live page so a compatible browser agent can use the product's own operational contract.

Challenges we ran into

The hardest part was making agent actions and visible UI state feel like one system instead of a chat demo layered on top of a dashboard. We also had to model advancement safely: status checks must be useful and read-only, while actual transitions remain explicit, auditable, and governed by published rules.

Designing a workflow builder that stays flexible across study types was another challenge. We avoided study-specific terminology and made stages, visits, requirements, and gates configurable.

Accomplishments that we're proud of

We built a polished multi-study workspace where fourteen WebMCP actions operate on the same state people see. The challenge demo shows retrieval, checklist creation, synchronized completion, audit history, and a guarded workflow decision in under three minutes.

What we learned

WebMCP is most compelling when tools represent meaningful domain actions rather than page-navigation shortcuts. Clear schemas, workflow provenance, read/write separation, and visible state synchronization make an agent experience more trustworthy and useful.

Judge demo

1. Ask: What does SUB001 need today?
2. Ask: Create SUB001's visit checklist.
3. Ask: Mark task T-001-3 complete for SUB001.
4. Ask: Can SUB001 advance to the next stage?

The final check is intentionally blocked and explains the remaining tasks, assessment, and samples. If the demo has already been used, choose Reset challenge demo in the Agent workspace.

What's next for ProtocolOS

Next, ProtocolOS can add protocol import, reusable workflow templates, role-based approvals, integrations with validated research systems, and configurable audit exports—while preserving the same human-visible, workflow-governed approach.

Built With

• openai
• react
• tailwind