Context Mode – 315 KB of MCP output becomes 5.4 KB in Claude Code
Context-aware summarizer cuts Playwright dumps from 56KB to under 1KB with zero friction.
Stops copy-pasting stack traces by piping browser errors and terminal logs directly to Claude.
Developers using Claude Code or similar AI coding assistants
Aider · Continue · Cursor
I am making this tool to make interaction gathering easier. It records user interaction in browser, make a prompt that describes what happened, send the events to a local MCP server that Claude can access. A command line tool that streams terminal output, in my case, server log, to the same MCP so claude can access both what happened in browser and terminal as a stream of events.
The idea is to emphasize that ideas often come from product, in contrast to mainstream belief that idea leads to product. The end is the start. Hence the project name janus.
It is still very early in development so would love to hear some feedback.
Context-aware summarizer cuts Playwright dumps from 56KB to under 1KB with zero friction.
Auto-memory hooks for Claude Code, but context window bloat within sessions remains unsolved.
Captures the 'why' behind AI changes live, fixing git blame's inability to show reasoning.
Ditch Usersnap: open-source widget feeds DOM context to AI agents via MCP.
First MCP security scanner addressing early Docker Hub moment—but audience is still nascent.
Structured pattern library beats noisy memory for LLM agents.