Tools

Astronomer's official data engineering plugin: write Airflow DAGs to best practices, debug pipeline failures, trace data lineage, profile warehouse tables and migrate from Airflow 2 to 3, with a built-in Airflow MCP server and session hooks.

Research guardrails for PhD students: reproduce arXiv papers, debug training anomalies with evidence, compare runs at matched epochs, check paper-code consistency, verify BibTeX, find literature gaps and predict reviewer comments. No MCP or extra dependencies; mostly for ML research.

Browser automation MCP built on Selenium WebDriver for Chrome, Firefox, Edge and Safari: locate elements, click and type, upload files, take screenshots, run scripts, and handle alerts and cookies. A good fit for test teams already using Selenium.

Official FinLab quant skill: lets agents use FinLab data (mainly Taiwan stocks, plus US, Japan, Korea and Hong Kong) to write stock-picking strategies, run backtests and analyze factors. Needs a FinLab account (free tier has a daily data cap), Python 3.10+ and uv. Not investment advice.

Lets agents run local ComfyUI workflows to generate images, audio and video, regenerate from earlier results with new parameters, check the queue and switch models. Install and start ComfyUI, then clone this repo and run the server with Python (port 9000 by default).

Official Databricks plugin with skills for the CLI and auth, Genie data queries, Apps, Lakebase, Lakeflow jobs and pipelines, Model Serving and DABs deployment, plus hooks for prompt routing, session context and auth failure hints. Needs a Databricks account.

Official Databricks skills guiding agents to build Lakeflow pipelines and jobs, Apps, Lakebase, Vector Search, Model Serving and AI/BI dashboards on the lakehouse. Needs the Databricks CLI.

Official testing skills from TestMu AI (formerly LambdaTest) covering 40+ frameworks including Selenium, Playwright, Cypress, Appium, pytest and JUnit, guiding agents to write unit, end-to-end and mobile tests. Runs locally; cloud cross-browser and real-device runs need an account.

radare2's official MCP server, written in C on the native r2 API: agents open binaries to disassemble, decompile and analyze them, with read-only mode, sandboxing and connections to remote r2 sessions. Install radare2 first, then r2mcp via r2pm.

Apache Doris's official MCP: gives agents read-only access to explore table metadata, run and explain SQL, diagnose slow queries, and check cluster status, load jobs and data governance info. Needs Python 3.12+ and Doris 2.0 or later.

Offline translation sync for coding agents: hands only new strings in JSON, YAML, PO, CSV and other locale files to the agent for translation and never overwrites human-edited translations, saving tokens. Needs Node.js; calls no translation API itself.

ComfyUI's official MCP: lets agents drive local ComfyUI image and video workflows, list installed nodes, models and templates, submit jobs and fetch results, and start or stop ComfyUI. Needs a machine that can run ComfyUI (usually a GPU) and comfy-cli; partner models consume Comfy credits.

QuantConnect's official CLI drives the open-source LEAN engine locally or in the cloud to download data, backtest and optimize quant strategies and deploy them live. Requires a paid-tier QuantConnect organization; strategies in Python or C#; local runs need Docker. Get expert review.

Postman's official MCP server: lets agents manage workspaces, collections, environments and API specs, help with API testing and generate client code from API definitions. The remote version uses OAuth; the local version needs a Postman API key.

Crowdin's official CLI: push source files and pull translations back in place, track progress, manage glossaries and TM, and run machine or AI pre-translation, with JSON output for agents. Needs a Crowdin account (free trial; free for open source) and a token; crowdin init sets up config.

Cross-platform UI automation skills built on Midscene.js visual recognition: control browsers, desktops, Android, iOS and HarmonyOS devices in natural language, and set up Vitest + Midscene E2E tests. Bring your own vision model API key; maintained by the Midscene team.

Hugging Face's official MCP server: lets agents search models, datasets and Spaces on the Hub and call thousands of Gradio AI apps directly. Hosted remotely; sign in with your Hugging Face account or authorize with a token.

Query and manage Microsoft 365 and Azure tenants in natural language via Microsoft Graph: create users and security groups, check devices and Conditional Access policies and view cloud costs. Needs a work account with the right permissions; changes apply immediately.

Grafana Labs' official skills: dashboards and alerting, PromQL, Loki, Tempo, Mimir and Pyroscope queries, Prometheus cardinality troubleshooting, Grafana Cloud management and cost optimization, k6 load testing and plugin development. Some skills need Grafana Cloud.

The Dart team's official MCP server (dart mcp-server): analyze project errors, search pub.dev and manage dependencies, connect to running Flutter apps for hot reload, and read runtime errors and the widget tree. Needs Dart 3.9+; still experimental.

Alibaba Cloud's official agent skills: ECS diagnosis and DR, RDS/PolarDB database checks, PAI training and EAS deployment, ACK clusters, RAM permissions, OSS and network troubleshooting, and Model Studio. Needs an AccessKey or aliyun CLI credentials.

Firecrawl's official plugin: brings the Firecrawl CLI in as skills so Claude Code can search the web, scrape pages to Markdown, map site links or crawl whole sites. Install firecrawl-cli and set an API key first; self-hosting supported.

Renders a screen reader spec for a component directly in Figma: focus order plus per-state VoiceOver (iOS), TalkBack (Android) and ARIA (Web) property tables and announcements. Part of uSpec; reads the component .md spec from create-component-md and needs a Figma MCP and a configured template key.

MinerU's official MCP: parses PDF, Word, PowerPoint and images into Markdown via the mineru.net cloud API, keeping tables and formulas. The keyless Flash mode works out of the box; add a token for higher quotas and more formats.