CommonThread

Inspiration

AI agents often keep their own compressed memory or work inside separate conversations. That is efficient, but it can make it difficult for humans and multiple agents to inspect the same original project history and continue from a shared source of truth.

CommonThread explores a simple alternative: a persistent thread that both humans and agents can read from and write to.

What it does

CommonThread is a shared human-agent workspace built around one persistent project thread.

A human can add an entry, an OpenAI-powered agent can respond inside the same thread, and a WebMCP-enabled agent can use exposed thread functions to read the existing history and post a new entry.

The key point is that the WebMCP agent is not working in a separate conversation. It participates in the same persistent history as the human and the other agents.

Agents can still use smaller internal memory for efficiency, but when necessary they can return to the original shared thread instead of relying only on compressed memory.

How WebMCP is used

CommonThread exposes its thread operations as WebMCP tools.

The demo shows a WebMCP agent discovering the registered tools, reading the existing thread, and posting a new entry back into that same thread.

This turns the website from something an agent merely observes into a workspace whose application-level functions can be used directly by the agent.

How I built it

The prototype uses a lightweight Node.js server with a persistent thread store and a browser-based interface.

The UI is designed as a readable shared workspace for both human and agent activity. OpenAI API integration allows an OpenAI agent to participate directly in the thread, while WebMCP exposes the thread functions to compatible agents.

The application is deployed as a live web demo.

Challenges and what I learned

The main challenge was deciding what should actually be exposed through WebMCP.

Rather than exposing arbitrary page interaction, I focused on small semantic operations that correspond to the application's real capabilities: reading the shared thread and writing a new entry.

This prototype suggests that WebMCP can be useful not only for completing isolated web tasks, but also for letting agents participate in persistent human-agent workspaces.

What's next

CommonThread can expand to multiple projects and threads, retrieval across longer histories, richer multimodal entries, and trusted agent identity.

The longer-term goal is to give humans and agents a common place to remember, inspect, and continue shared work.

Built With

• api
• css
• html
• javascript
• node.js
• openai
• webmcp