wyd

• 

  wyd tui

• 

  wyd webmcp

Inspiration

Coding agents often leave browsers, MCP servers, dev servers, and other processes running after a session ends.

ps tells me what is running, but not who started it or whether it is safe to remove. I built wyd to answer that.

What it does

wyd connects running resources to the coding-agent session and project that created them.

For this challenge I added wyd web and six WebMCP tools. A browser agent can find ended sessions, inspect leftovers, explain process provenance, focus resources in the UI, and prepare a cleanup proposal.

Cleanup itself is intentionally left to the human.

How I built it

wyd is mostly Rust. The web dashboard uses the same runtime/provenance engine as the CLI and TUI.

WebMCP tools are registered with document.modelContext.registerTool, and agent actions update the same UI state the human sees.

The hosted demo uses deterministic synthetic data.

Challenges

The main challenge was deciding what the agent should be allowed to do.

I separated investigation from execution: WebMCP can explain and prepare cleanup, but cannot perform destructive actions.

I also separated ownership from lifecycle state. A process can belong to an ended session and therefore be both confidently owned and a leftover candidate.

What I learned

Structured tools work much better than asking an agent to scrape a UI.

The useful pattern became: agent investigates → UI reflects it → human decides

Built With

• codex
• css
• docker
• github-actions
• html
• javascript
• linux
• macos
• model-context-protocol
• rust
• webmcp