Hi HN — I just shipped the first feature update for Spring AI Playground since it became an official Spring AI Community incubating project.
The goal is to treat MCP tools as runtime entities you can build, inspect, and iterate on live—without rebuilds or redeployments.
What’s in this update:
- No-code / low-code Tool Studio: create AI-callable tools in the browser using JavaScript (ECMAScript 2023) and run them sandboxed in the JVM (GraalVM Polyglot).
- Live built-in MCP server: tools are evaluated and registered dynamically to an embedded MCP server, becoming immediately available (no restart).
- MCP inspection & debugging: inspect registered tools, schemas/parameters, and execution results to understand agent behavior.
- Agentic chat: test end-to-end flows that combine LLM reasoning, MCP tools, and optional RAG context in one UI.
It also ships with working example tools meant to be copied/modified as templates:
- googlePseSearch - Google PSE web search (API key)
- extractPageContent - Extract clean text from URLs (RAG prep)
- buildGoogleCalendarCreateLink - Generate calendar add links
- sendSlackMessage - Slack webhook messaging
- openaiResponseGenerator - OpenAI API calls (API key)
- getWeather - wttr.in weather
- getCurrentTime - ISO-8601 time
Everything runs local-first with Ollama by default (OpenAI-compatible APIs supported). No cloud required.
Local-first by default using Ollama and OpenAI-compatible endpoints/config, so it can run without required cloud services.
GitHub:
https://github.com/spring-ai-community/spring-ai-playground
Question for folks building with MCP/agents: what inspection/debugging workflows do you wish every MCP client/UI supported (e.g., clear schema views, tool-call tracing, reproducible runs, etc.)?
