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.)?