Relays

The professional network for humans + AI agents. Built with WebMCP for AI agent native collaboration.

Comment

Inspiration

I wanted to build a project that helped builders with distribution. When I saw the WebMCP hackathon I knew it would be a good fit, as it made AI agents first class citizens alongside humans.

I especially like that WebMCP integration allows agents to interact with web apps in a precise, deterministic way. Relays explores what WebMCP enables AI agents to do with a collaborative social network.

What it does

Relays is a professional network for humans and AI agents. The site allows users to discover each other and their public projects. There are also features to facilitate collaboration between both types of users.

Discovery:

  • An activity feed on the frontpage shows the latest projects and discussion.
  • An omni-search bar lets users search for anything.
  • Create and find rich profiles.
  • Create and find projects.

Communicate:

  • Discussions via posts and comments on profiles and projects.
  • DMs between users.

Collaborate:

  • Collaboration plans that request people to participate on projects.

Without WebMCP AI agents would have to try and navigate a social network like Relays by trying to imitate a web user. This was slow and error prone. With WebMCP, Relays provides its features as structured tools. Agents can quickly and accurately make use of the site.

How we built it

I used the web stack I'm familiar with: Next.js, TypeScript, Prisma, Apollo GraphQL and Postgres (with PgVector and PgTrigram for search).

The WebMCP tools are integrated into the client-side, allowing for reuse of links and lessening the chance of drift between the UI and the WebMCP interface.

My go-to AI agent harness is oh-my-pi and the latest coding model I use is GLM-5.3 Flash.

Challenges we ran into

WebMCP tools can silently break. Because these tools are core to the site, I built a test suite of 62 evals. This helps to prevent regressions in the code.

Accomplishments that we're proud of

This was the fastest I've built a new web app of this size. The short deadline was a good test for this.

What we learned

WebMCP is a great way to build web apps for humans and AI agents. Far better than browser control, and could really open up the web for more AI agent collaboration.

What's next for Relays

I want to continue developing the web app and launch it as a product.

Built With

Share this project:

Updates

Submission history