Agentic Shift Search

Inspiration

We are a team of marketers, developers and AI builders at Nursa striving to make our company not only visible to LLMs but also seamless for both humans and AI agents to interact with. We achieve this by structuring our content and data, implementing relevant schema markup, and actively exploring Model Context Protocol (MCP) integrations. WebMCP felt like a perfect enabler for making interactions with AI agents quicker, smoother, and completely frictionless. We couldn't wait to try it.

What it does

It allows agents to interact with the API endpoint including available clinician shifts on nursa.com. Instead of searching the shifts manually with the UI, we allow agents to use all our filters inside of WebMCP tools:

search_shifts - searches by license, state, specialty, minimum hourly pay, date range, and free text; returns up to 10 compact results. Data only - it never touches the list on screen.

apply_filters - takes the same filter inputs and writes into the shared filter store, so the visible list re-renders with the agent's choice. This is the human+agent moment the project is built around.

How we built it

We created a simple page displaying shifts available on our website and allowing for a filter selection. We added a script that registers WebMCP tools and references our endpoint. We tested it with WebMCP Model Context Tool Inspector

Challenges we ran into

1) WebMCP is really new. Right now it only works in ChatGPT's desktop app, and it doesn't behave the same way twice. The same prompt might work on one try and fail on the next. We spent a surprising amount of time just rewording tool descriptions until the model used them the way we wanted. 2) Dates gave us the most trouble. First a timezone bug: someone asked for shifts "from Sep 7th" and the filter got set to Sep 6th. Then ChatGPT told a user that shifts were "in 2027", because our responses only said "Sep 5" with no year and it guessed. We ended up sending full timestamps with the year and timezone, and kept the pretty labels just for display. 3) The AI would answer with stale results. Asking "RN shifts" and then "RN shifts in California" sometimes returned the first answer again, because the second search didn't properly reset the state from the first one.

Accomplishments that we're proud of

Quick implementation of a powerful feature we can later roll out throughout our platform.

What we learned

We weren't familiar with WebMCP before. Building this app gave us hands-on experience with its architecture, tool registration, and local testing workflows, giving us a clear view of both its massive potential for agentic search and its current limitations.

What's next for Agentic Shift Search

Further improvements to existing tools like expanding them with shift details, adding new tools and potential rollout to our main search pages, e.g. https://nursa.com/jobs/shifts

Built With

• json
• typescript
• webflow
• webmcp