Raindrop Workshop

Raindrop Workshop is a local debugger for AI agents. It runs as a daemon on your machine and streams each agent run into a browser interface as it happens: tokens, tool calls and spans arrive live rather than on a refresh. Running /instrument-agent inside a coding agent adds Raindrop tracing to the repository and opens the interface.

Beyond viewing traces, Workshop is designed for a coding agent to drive. It registers an MCP server named workshop, so a coding agent reads the captured traces, writes evaluations against the code in the repository, and iterates: write an evaluation, run the agent, observe the failure, change the code, re-run until the assertions pass. A separate /setup-agent-replay command scaffolds an HTTP endpoint that replays a production trace against the real agent code.

Traces are kept in a local SQLite database, and both the database path and the interface port are configurable. The raindrop workshop subcommands start, check, reset and update the local install.

The same raindrop binary connects a project to Raindrop Cloud, the hosted production-observability product, with no local daemon involved. A cloud setup command signs in, writes an organization write key to the project's environment file, and installs a hosted MCP server plus the raindrop-setup and raindrop-investigate skills into your coding agents, and a matching uninstall command reverses that without touching the local install. Local Workshop and the cloud install use distinct MCP server names and separate registries, so both can be present at once.