Raindrop launches Workshop: a local debugging tool for AI agents with self-healing eval loops
By
Alexis Gauba
Sketches a bagel but doesn't bake it through.
Summary
Raindrop introduces Workshop, a debugging tool for AI agents that provides a local UI and MCP (Model Context Protocol) for streaming traces, replaying agent runs with edited prompts/models/tools, and creating self-healing eval loops. It aims to solve the fragmented debugging experience where traces, harnesses, and eval suites are scattered across different platforms.
Key quotes
· 3 pulledYour agent fails at 1am, traces are in some SaaS dashboard, harness is on your machine, eval suite is in a third place, and Claude sees none of it.
Workshop is the first sane way to debug your agent locally.
The MCP lets you create self-healing eval loops. Claude Code reads your traces, writes evals, and fixes what's broken.
You might also wanna read
Workshop: How to Build Your Own Coding Agent
This is a workshop tutorial that teaches participants how to build their own coding agent, similar to tools like Roo Code, Cline, Amp, Curso
ghuntley.com·9mo ago
claude-replay: Convert AI Coding Agent Sessions into Interactive HTML Replays
claude-replay is an open-source tool that converts AI coding agent session logs (from Claude Code, Cursor, Codex CLI, Gemini CLI, and OpenCo
github.com·2mo ago
PiClaw: Self-Hosted AI Workspace with Multi-Provider LLM Support and Built-in Development Tools
PiClaw is a self-hosted AI workspace that packages the Pi Coding Agent into a comprehensive development environment. It features a streaming
github.com·2mo ago