ツール

Measurement-first audits and optimization of web quality using Lighthouse, Chrome DevTools performance traces and CrUX data, covering Core Web Vitals, WCAG 2.2 accessibility, SEO and best practices, for any frontend framework. By Addy Osmani (not an official Google project).

Open-source, agent-friendly BI: define metrics and dashboards in YAML and SQL, and rill init generates project instructions and MCP config for Claude Code so agents can build business dashboards and answer questions in natural language. Local use is free; publishing to Rill Cloud needs an account.

Postgres rules ranked by impact: query optimization, indexing, connection pooling, RLS security, schema design and lock troubleshooting, with wrong-vs-right SQL examples, applicable to any Postgres. Maintained by Supabase.

Unified search across 20+ academic sources including arXiv, PubMed, bioRxiv, Semantic Scholar and OpenAlex, returning deduplicated, normalized results and downloading full text via open-access channels first. Most sources need no API key.

Bright Data's official MCP: web search (Google, Bing, Yandex), any page to Markdown, structured data extraction from e-commerce, social and other platforms, and remote browser control. Needs an API token; includes 5,000 free requests a month.

Simon Willison's website screenshot CLI: batch screenshots from YAML, record demo videos, export PDF and HTML, run JavaScript in pages to extract data, and dump accessibility trees. Built on Playwright; install a browser on first use.

Runs open image models like FLUX, Qwen Image and Z-Image locally on a Mac: text-to-image, reference-based edits, inpainting and upscaling, all on-device. Needs Apple Silicon and uv; the first run downloads models and is slow. Check each model's license before commercial use.

Four tools for real-time web search, page content extraction, site mapping and full-site crawling, available as a remote server or run locally. Maintained by Tavily; needs an API key or OAuth sign-in, and a free account is available.

Uses hooks to enforce test-driven development in Claude Code: blocks implementation without a failing test, prevents code beyond what tests require, and can use lint results to push refactoring. Needs Node.js 22+. Still maintained, but the author suggests its successor Probity for new projects.

MiniMax's official CLI, built for agents: chat with MiniMax models from the terminal and generate images, video and speech, plus speech transcription, image understanding and web search. Works with the international platform. Requires a MiniMax Token Plan subscription.

A CLI and Python library by Simon Willison: imports JSON and CSV straight into SQLite with automatic table creation, runs in-memory SQL queries over files, and supports full-text search and table transformations.

Self-hostable long-term memory for agents, offering MCP, a REST API and a web dashboard: semantic search, knowledge graph and automatic memory consolidation. Embeddings are computed locally; choose SQLite, Cloudflare or Milvus as the backend.

Recommends daily papers from arXiv and Semantic Scholar by research interest, reads them and writes illustrated Obsidian notes and a paper graph; can also search venues like CVPR and NeurIPS. Needs Python 3, Obsidian, a vault path and an interests config; the web app needs an Anthropic API key.

MathWorks' official server that lets agents start your local MATLAB, write and run scripts and unit tests, and check code style and potential errors, for computing, simulation and data analysis. Needs a licensed MATLAB R2021a+ and the downloaded server binary.

Uses the Qdrant vector database as a semantic memory layer for agents: store information and retrieve it by meaning, connecting to a remote instance or a local path, with FastEmbed generating embeddings by default. Maintained by Qdrant.

Connects LLMs to ROS and ROS 2 robots: publish and subscribe to topics, call services and actions, set parameters, and read sensor and robot state, without changing robot code. The robot side needs rosbridge running.

Lets Claude Code watch videos: extracts frames with ffmpeg and transcribes timestamped audio from local files or YouTube links, to review footage, summarize lectures or find on-screen text. Needs Node.js 20+ and ffmpeg (yt-dlp for YouTube); transcribe with local Whisper or a Gemini/OpenAI key.

Gives agents controlled access to MySQL: list databases and tables, inspect schemas and sample data, and run SQL, with multi-database mode, SSL and SSH tunnels. Configure your own database user, ideally with least privilege.

Pulls new papers from Hugging Face Daily Papers and arXiv, scores them by research area into a recommendation page, and writes deep-read Obsidian notes for key ones; can also batch-read Zotero collections. Needs Python 3, Obsidian and poppler; copy the whole skills folder per the README.

LangChain's official skill collection: framework selection, quickstarts, middleware, persistence, human-in-the-loop approval, RAG and eval design across LangChain, LangGraph and Deep Agents, in Python and TypeScript. Also installable as a Claude Code plugin. Still in early development.

Connects agents to live Jupyter notebooks: read, write, insert and execute cells and read multimodal outputs like charts, for interactive data analysis. Bring your own JupyterLab or JupyterHub; open source, maintained by Datalayer.

Lets agents build Xcode projects and drive the iOS Simulator: 29 scripts cover building and testing, semantic navigation and taps via the accessibility tree, screenshot diffs and device lifecycle, with compact token-saving output. Needs macOS, Xcode and idb.

MathWorks' official toolkit for reading, building, modifying and testing Simulink models via the MATLAB MCP server, with skills for control design, Simscape, code generation and compliance checks. Needs licensed MATLAB R2023a+ and Simulink; run the official installer in MATLAB first.

The standard machine translation scorer: computes BLEU, chrF and TER with significance tests and downloads public test sets such as WMT, to compare engines or prompts. Needs Python 3.9+ and reference translations; automatic metrics are a rough screen, so final copy still needs human review.