Brief — Research with your agent

Inspiration

Chat is a terrible place to hold research. Claims, sources, and caveats scroll away into mush. We wanted a board where a person and ChatGPT could see the same brief — not a summary pasted back into the thread, but a living evidence map they edit together.

WebMCP makes that possible: the site exposes tools to add claims, link sources, flag contradictions, and export, on the page you’re already looking at.

What it does

Brief is an agent-native evidence board:

Set a research question and title
Add claims with stance (supports / neutral / disputes) and confidence
Attach sources with snippets and link them to claims
Flag contradictions and pin notes
Summarize and export the brief
Use a full human UI or ChatGPT via WebMCP tools on the same Zustand store
Inspect tools + activity feed; run ?demo=1 for the happy path

How we built it

Vite + React + TypeScript SPA (base: '/brief-webmcp/')
Zustand shared store for UI and WebMCP execute handlers
18 site tools with readOnlyHint on reads; clear_board / export_brief as side effects
Feature-detect document.modelContext → navigator.modelContext (top-level page)
MIT, static GitHub Pages deploy — Built with Codex

Challenges we ran into

Keeping agent writes and human edits in lockstep; marking consequential tools correctly; designing a board that’s scannable in a <3 minute demo without becoming a toy.

Accomplishments that we're proud of

A third WebMCP submission in a new lane (research, not shopping or canvas): claim→source→flag loops, summarize/export, and a judge-friendly tools panel.

What we learned

WebMCP’s superpower is co-presence. Research needs shared structure more than another chat answer — Brief treats the agent as a colleague on the board, not a ghostwriter in the sidebar.