Signal Desk Workboard

Inspiration

This project began with a practical problem: important operational work often starts in WhatsApp. A customer message, an incident report, or a request for help arrives in a chat, then someone needs to understand it, assign it, follow up, and record the outcome.

We had already been exploring a local WhatsApp ingestion workboard. However, that system relies on private authentication and persisted chat/media data. It is intentionally local-only: publishing it would expose sensitive data and would not be a reliable fit for a serverless deployment.

That constraint became the inspiration for Signal Desk.

Instead of pretending that private WhatsApp data should be pushed to the public web, we built a safe, fictional, browser-only workboard that represents the workflow after a message becomes an operational case. It lets us demonstrate the meaningful question: how can people and AI agents work on the same case safely?

What it does

Signal Desk is a fictional support workboard where a person can create, review, edit, and move cases through New, Triage, Investigating, and Resolved.

The important difference is that the page also exposes focused WebMCP tools. An agent can list cases, open a case, create a new case, update its record, and propose a stage move using structured site capabilities rather than guessing from screenshots or clicking around blindly.

The human remains the decision-maker. When an agent wants to create, edit, or move a case, Signal Desk creates a visible pending action instead of silently changing the board. The person can approve or cancel it. After approval, the agent reads the actual board state back to verify the outcome.

This means the product supports both kinds of work:

Humans can use the normal visual interface.
Agents can use WebMCP to work through the same case model.
Important changes remain visible, reversible before approval, and verifiable afterward.

How we built it

We separated the original private WhatsApp concept from the deployable demonstration.

The original ingestion idea stays local, where private authentication and persisted conversations belong. Signal Desk is the public, safe companion: it uses fictional data that lives only in the browser and can be deployed to Vercel without exposing real messages, media, credentials, or customer information.

We built a case workboard with rich case details, notes, formatting, case creation, manual editing, stage movement, and a visible activity history. We then implemented WebMCP tools for the same operations.

For state-changing tools, we added an approval workflow:

The agent proposes an action.
The website returns an awaiting approval state and an approval ID.
A person can approve or cancel the specific action.
The agent confirms the approved action.
The agent reads back the resulting case or board state before claiming success.

Challenges we faced

The biggest challenge was not making an agent able to act. It was making agent action trustworthy.

A tool that can move a case directly is convenient, but it can also hide a consequential decision. We designed the approval state so that creating a case, changing its content, and moving it to Resolved are explicit, separate decisions.

We also learned that an approval alone is not enough. The agent must verify the result after the action completes. During testing, we improved the workflow so the agent does not merely report “done”; it checks the updated record or board state and reports evidence of the outcome.

What we learned

WebMCP is valuable because it gives an agent a clear contract with a webpage. The agent does not need private backend access, fragile screen automation, or permission to control everything. The website declares the useful actions, their inputs, and their safety boundaries.

Signal Desk shows how an ongoing, message-driven workboard project can become agent-interactive and deployment-ready without exposing the private WhatsApp layer behind it.

The goal is not to replace people with automation. It is to turn scattered reports into shared, structured work—where agents can help move work forward and people retain authority over the decisions that matter.