Inspiration
What it does
How we built it
Challenges we ran into
Accomplishments that we're proud of
What we learned
What's next for Sigmora Live Relay
Inspiration
Going live is when creators most need cross-platform promotion—and when they can least afford to leave the stream to edit, rewrite, upload, and check posts. Sigmora Live Relay turns one selected live event into a reviewable 12-second vertical campaign on a shared release board.
What it does
The page exposes eight narrow WebMCP tools. An agent reads the visible brand and destination context, finds or declares the source live event, drafts the relay, polls generation, revises copy, queues exact destination accounts, releases the human-approved revision, and checks each destination independently. Every tool call updates the same board the creator sees.
The consequential checkpoint remains human. Queue items are bound to exact account IDs and one immutable revision. The creator reviews those items in the UI and approves them there; approval is deliberately not exposed as a tool. Any revision invalidates the receipt. Release requires that receipt plus an idempotency key, and partial failure can never be reported as total success.
Why WebMCP
This is a strong WebMCP fit because the useful context and collaboration live on the open page: current creative state, selected destinations, approval, and receipts. Without WebMCP, an agent must infer buttons and duplicate state. With WebMCP, it uses an explicit contract while the person keeps visual control.
People and agents can now share one live workspace: the agent handles multi-step coordination while the creator remains present with the audience and retains authority over publication.
How we built it
The challenge implementation is framework-free TypeScript built around document.modelContext.registerTool(...), closed JSON Schemas, typed result envelopes, an observable shared controller, page-scoped teardown, a strict HTTP adapter, and a deterministic local provider. The normal manual interface continues to work when WebMCP is absent.
The eight tools cover context, source discovery, drafting, job polling, revision, exact destination queuing, approved release, and independent status verification. Unit and end-to-end tests exercise schemas, lifecycle cleanup, revision invalidation, idempotency, partial failure, and the manual fallback.
Challenges
The hardest design problem was preserving human authority without making the agent workflow brittle. We solved it with an immutable approval receipt bound to an exact revision and destination set, plus clear state transitions visible to both sides. Another challenge was truthful partial failure: each destination receives an independent receipt, so a mixed outcome stays mixed.
What we learned
WebMCP is most compelling when it is not simply a hidden automation API. The shared page should remain the source of truth, and the tool contract should expose exactly the capabilities that benefit from agent reasoning while keeping consequential consent visible and human.
What's next
The public reference is intentionally simulated and says so throughout the UI and tool results. Next we would connect the strict provider adapter to real publishing APIs, preserving the same approval, idempotency, readback, and partial-failure guarantees.
Log in or sign up for Devpost to join the conversation.