ツール

Unbabel's machine-translation quality metric: comet-score rates translations per sentence and overall, XCOMET marks error spans and severity, and CometKiwi works without a reference. Good for spot-checking MT. Needs Python 3.8+; models download from Hugging Face, some need a license; GPU helps.

A research plugin for Claude Code: commands like /omp:survey, /omp:experiment and /omp:write chain literature review, topic selection, experiments, writing and self-review, with 5 specialized agents and 34 research skills that remember project progress across sessions. Experiments lean toward ML.

The five-agent PaperOrchestra pipeline (outline, figures, literature review, section writing, polishing) as skills: turns research ideas, experiment logs and a conference template into a LaTeX paper and checks citations via Semantic Scholar. Needs Python; verify the draft yourself.

Official SonarSource server: connects to SonarQube Server or Cloud so agents can query code issues, quality gates, security hotspots and dependency risks, and analyze file contents directly. Needs a SonarQube token; Docker is the recommended way to run it.

Firecrawl's official CLI: scrape, search, map and crawl sites and interact with pages in a browser from the terminal, with Markdown output; can also install companion skills for coding agents. Needs a Firecrawl account or API key, or point it at a self-hosted instance.

Official Firecrawl skill: drives the Firecrawl CLI to search, scrape, map site URLs, batch-crawl, interact with pages and monitor changes, returning LLM-ready Markdown. Search and scrape have a rate-limited free tier; crawling and monitoring need an account or self-hosted instance.

Redis's official MCP server: read and write strings, hashes, lists, streams, JSON and other data structures in natural language, with vector indexing and search and pub/sub. Bring your own Redis instance; runs via uvx.

Unity's official plugin bundling skills for game development, in-app purchases and ad monetization, and performance optimization, and can drive an open editor through the Unity CLI. Supports Unity 6+, installs in Claude Code and Codex; currently in beta.

Lets agents deploy a local folder or file contents straight to Google Cloud Run, list services and view their details and logs, and list or create GCP projects. Maintained by Google; sign in with gcloud and set up Application Default Credentials first.

Official Docker skills: containerize projects, optimize and harden Dockerfile builds, orchestrate with Compose, use Docker Sandboxes and Docker Agent, with confirmation guardrails before deletes and other destructive actions.

Railway's official CLI: deploy with railway up and manage infrastructure config, with token auth for non-interactive CI runs. The installer can set up the Railway skill and MCP for Claude Code, Codex and others in one step. Needs a Railway account.

Elastic's official skills teaching agents to use Elasticsearch and Kibana correctly: index design, ES|QL queries, search relevance tuning, dashboards, observability troubleshooting and security alert triage. Currently a technical preview.

Manage Argo CD in natural language: list and inspect applications, create, update and delete them and trigger syncs, and view resource trees, workload logs and events. Maintained by Argo Project Labs; needs your Argo CD URL and an API token, with an optional read-only mode.

Chroma's official MCP: lets agents create and manage collections, add documents, and retrieve them with vector search, full-text search and metadata filters. Connects to in-memory or local persistent stores, self-hosted instances or Chroma Cloud.

Official ElevenLabs skills: text to speech, speech to text, sound effects and music generation, voice changing, vocal isolation, multilingual dubbing and building voice agents. Needs an ElevenLabs API key.

MCP server for the Lean 4 theorem prover: lets agents read proof goals, diagnostics and definitions and search theorems via LeanSearch and Loogle, helping mathematicians and students with formal proofs. Needs uv and the Lean toolchain; run it from the Lean project root.

Google's official security MCP servers for Google SecOps (Chronicle), SecOps SOAR, Google Threat Intelligence and Security Command Center: investigate alerts, hunt threats, look up IOCs and review cloud risks. Needs Google Cloud credentials or a VirusTotal API key.

68 trading and quant skills focused on crypto and DeFi: market and on-chain data, technical indicators, backtesting, position sizing and risk, prediction markets, and tax tools for cost basis and wash sales. Most need Python and data API keys (often free tiers); get expert review.

Lets agents query Prometheus metrics with PromQL and browse metric metadata and scrape targets to help investigate alerts and performance issues. Supports basic auth, bearer tokens and multi-tenancy; Docker is the recommended way to run it.

Appium's official MCP server: lets agents start sessions on Android and iOS emulators and real devices, locate elements, tap, type and take screenshots, and generate Java/TestNG test code. Needs the Android SDK or Xcode.

Official Tavily skills: let agents search the web, extract page content, discover site URLs, crawl in bulk and run cited deep research through the Tavily CLI (tvly). Install the CLI first; search and extract have a limited no-login tier, other features need sign-in or an API key.

Lets agents work with Salesforce orgs authorized on your machine: SOQL queries, metadata deploy and retrieve, Apex test runs and permission set assignment, plus toolsets for LWC development. Maintained by Salesforce; authorize your org with the Salesforce CLI first.

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.