Raindrop Workshop GitHub Repository

Raindrop Workshop is a debugger that shows every token, tool call, and span an agent generates locally in real time. It connects coding agents like Claude Code so they can read traces, write evals against the codebase, find problems, and fix them.

Installation is a single line: curl -fsSL https://raindrop.sh/install | bash. To run it directly from source, run git clone followed by bun install and bun run dev. This starts both the local Workshop daemon and the Vite UI together, viewable at http://localhost:5899.

On the agent side, run /instrument-agent to attach Raindrop tracing and open Workshop in your browser. Traces are then streamed to the UI instantly, with no manual refresh or polling needed.

The core features are as follows.

• Live streamed traces: Tokens, tool calls, and spans are reflected in the UI as soon as they occur.
• Coding-agent integration: Claude Code reads traces, writes evals, and then fixes failures.
• Self-healing eval loop: It cycles through writing evals, running the agent, checking failures, fixing code, and rerunning until all assertions pass.
• Local replay: /setup-agent-replay creates an HTTP endpoint that replays production traces against the actual agent code.

The scope of support is also broad. It supports TypeScript, Python, Go, Rust, and is compatible with Vercel AI SDK, OpenAI Agents SDK, Anthropic SDK, Claude Agent SDK, LangChain, LangGraph, CrewAI, Mastra, Pydantic AI, DSPy, Google ADK, Strands, Agno, and Deep Agents. Runtime environments include AWS Bedrock, Azure OpenAI, and Vertex AI, and supported coding agents include Claude Code, Codex, Devin, Cursor, and OpenCode.

Configuration is controlled via RAINDROP_WORKSHOP_PORT (default 5899), RAINDROP_WORKSHOP_DB_PATH (default ~/.raindrop/raindrop_workshop.db), and RAINDROP_LOCAL_DEBUGGER. The CLI includes raindrop workshop, raindrop workshop setup, raindrop workshop status, raindrop workshop reset, and raindrop update.