# MCP / Skills Integration Catalog > [繁體中文](./mcp-skills-catalog.md) | [简体中文](./mcp-skills-catalog.zh-Hans.md) | **English** > This is a task-based catalog for finding tools when needed, not an install list. Start with an official entry point; expand a category only when the work calls for it. ## 📌 How to use this catalog 1. First state whether you need to read data or write to a real service. 2. Prefer official hosted / reference entry points, then check maintenance, licensing, and permissions. 3. Start with test data, read-only access, and least privilege. Keep human approval before write, send, or delete actions. Installation and testing belong in [Stage 5](../stages/05-claude-code-ecosystem.en.md); this page only helps you find candidates. ## 🧩 Five terms to separate first - **MCP Server**: a program or hosted service that exposes data or actions as MCP tools. - **Skill**: reusable instructions, scripts, templates, and references; loading differs by host. - **Plugin**: a host-specific package that may bundle Skills, commands, hooks, or MCP configuration; it is not the MCP specification. - **Remote MCP**: an MCP Server operated by a provider, usually using OAuth, so you do not run it locally. - **Permission Boundary**: the range an Agent can actually read, write, send, or delete. A tool being capable does not mean every action should run. ## 📚 Five safe starting points | Starting point | Learn first | Editorial rating | |---|---|---| | [Official MCP Registry](https://registry.modelcontextprotocol.io/) | Find published MCP Servers; still check the maintainer, permissions, and source | ⭐⭐⭐⭐⭐ | | [modelcontextprotocol/servers](https://github.com/modelcontextprotocol/servers) | See educational reference implementations; they are not a production recommendation | ⭐⭐⭐⭐⭐ | | [Notion MCP](https://developers.notion.com/guides/mcp/overview) | See how a hosted OAuth MCP follows the user’s workspace permissions | ⭐⭐⭐⭐⭐ | | [GitHub MCP Server](https://github.com/github/github-mcp-server) | Learn OAuth, tool groups, and repository permissions from an official implementation | ⭐⭐⭐⭐⭐ | | [anthropics/skills](https://github.com/anthropics/skills) | See how Agent Skill folders, instructions, and resources fit together | ⭐⭐⭐⭐⭐ | > ⚠️ An MCP Server may touch real data. Even with an official source, confirm the signed-in account, scopes, writable tools, and approval screen. ### Index 1. [Notes / Knowledge Bases](#1-notes--knowledge-bases) 2. [Office Documents (Word / Excel / PowerPoint / PDF)](#2-office-documents-word--excel--powerpoint--pdf) 3. [Google Workspace](#3-google-workspace) 4. [Microsoft 365](#4-microsoft-365) 5. [Dev Collaboration (GitHub / Atlassian / Slack…)](#5-dev-collaboration-github--atlassian--slack) 6. [Databases](#6-databases) 7. [Browser Automation / Web Scraping](#7-browser-automation--web-scraping) 8. [Design (Figma / Excalidraw)](#8-design-figma--excalidraw) 9. [Monitoring / Observability](#9-monitoring--observability) 10. [Media / Streaming (YouTube / Spotify)](#10-media--streaming-youtube--spotify) 11. [Chinese-language Ecosystem](#11-chinese-language-ecosystem) 12. [Other Common (Cloudflare / Stripe…)](#12-other-common-cloudflare--stripe) 13. [Research Workflow Skills (Academic / Paper / Literature)](#13-research-workflow-skills-academic--paper--literature) 14. [Multi-LLM Delegation Skills](#14-multi-llm-delegation-skills) 15. [Finance / Trading Agents](#15-finance--trading-agents) 16. [Web Search / Retrieval](#16-web-search--retrieval) 17. [Security / MCP Governance](#17-security--mcp-governance) --- ## 1. Notes / Knowledge Bases

Expand selected entries for category 1

### [Notion MCP](https://developers.notion.com/guides/mcp/overview) ⭐⭐⭐⭐⭐ | Field | Value | |---|---| | Delivery | Notion hosted remote MCP; OAuth | | Rating | ⭐⭐⭐⭐⭐ (**official**) | **What it does**: Notion’s official hosted MCP can search, read, create, and update content the user can already access. **Audience**: People who use Notion for notes, projects, or a wiki and want an MCP-capable client. **Notes**: OAuth follows user permissions; review the approval screen before writing. The older open-source `makenotion/notion-mcp-server` is no longer actively maintained and is not a default for new installs. ### [MarkusPfundstein/mcp-obsidian](https://github.com/MarkusPfundstein/mcp-obsidian) ⭐⭐⭐⭐ | Field | Value | |---|---| | License | MIT | | Rating | ⭐⭐⭐⭐ (community maintained) | **What it does**: read/write your Obsidian vault via the Obsidian REST API community plugin. **Audience**: heavy Obsidian users wanting Claude Code to organize daily notes, auto-link, search across files. **Notes**: requires the [Local REST API](https://github.com/coddingtonbear/obsidian-local-rest-api) plugin in Obsidian. ### [PleasePrompto/notebooklm-skill](https://github.com/PleasePrompto/notebooklm-skill) ⭐⭐⭐⭐ | Field | Value | |---|---| | License | MIT | | Rating | ⭐⭐⭐⭐ | **What it does**: a Claude Code Skill that uses browser automation to query Gemini Notebook (formerly NotebookLM), with citation-backed answers. **Audience**: people who manage papers / research notes in Gemini Notebook (formerly NotebookLM) but want to query from Claude Code in one prompt. **Notes**: requires Google account auth. ### [teng-lin/notebooklm-py](https://github.com/teng-lin/notebooklm-py) ⭐⭐⭐⭐ | Field | Value | |---|---| | License | MIT | | Rating | ⭐⭐⭐⭐ | **What it does**: unofficial Gemini Notebook (formerly NotebookLM) Python API, CLI, and agentic Skill. **Audience**: people doing programmatic or batch operations in Gemini Notebook (formerly NotebookLM). **Notes**: unofficial; may break with Google policy changes — check the issue tracker before relying on it. ### [ergut/mcp-logseq](https://github.com/ergut/mcp-logseq) ⭐⭐⭐ | Field | Value | |---|---| | License | MIT | | Rating | ⭐⭐⭐ | **What it does**: read/write Logseq graph via Logseq's Local HTTP API. **Audience**: Logseq users automating daily journals, cross-page links, backlink queries. **Notes**: enable Logseq's HTTP API (Settings → Features → HTTP API). ### [skridlevsky/graphthulhu](https://github.com/skridlevsky/graphthulhu) ⭐⭐⭐ | Field | Value | |---|---| | License | MIT | | Rating | ⭐⭐⭐ (covers both Logseq + Obsidian) | **What it does**: a broad tool set across navigation, search, analysis, writing, journals, flashcards, whiteboards. **Audience**: people using both Logseq and Obsidian who don't want two MCP servers. **Notes**: community project; broad tool surface but each tool is relatively basic. ### [ankimcp/anki-mcp-server](https://github.com/ankimcp/anki-mcp-server) ⭐⭐⭐ | Field | Value | |---|---| | License | MIT | | Rating | ⭐⭐⭐ | **What it does**: create / query / batch-edit Anki decks via AnkiConnect. **Audience**: people using Anki for languages / medicine / law — let the LLM auto-generate cards from study material. **Notes**: requires Anki Desktop + the [AnkiConnect](https://ankiweb.net/shared/info/2055492159) addon. ---

## 2. Office Documents (Word / Excel / PowerPoint / PDF)

Expand selected entries for category 2

### [anthropics/skills](https://github.com/anthropics/skills) ⭐⭐⭐⭐⭐ | Field | Value | |---|---| | License | No license file (none provided upstream; confirm terms before use) | | Rating | ⭐⭐⭐⭐⭐ (**official examples**) | **What it does**: Anthropic's official Agent Skills examples include document workflows for docx, xlsx, pptx, and pdf files. **Audience**: people who want to study how a standard Skill folder, `SKILL.md`, scripts, and resources work together. **Notes**: this is a Skills collection, not MCP. Preloading, installation, and support differ across Claude / Agent surfaces; check the repository and current host documentation before use. ### [haris-musa/excel-mcp-server](https://github.com/haris-musa/excel-mcp-server) ⭐⭐⭐⭐⭐ | Field | Value | |---|---| | License | MIT | | Rating | ⭐⭐⭐⭐⭐ (community-maintained Excel MCP) | **What it does**: Excel file manipulation MCP — read / write / modify cells, formulas, sheets. **Audience**: people working with Excel reports daily who want LLM-driven data filling and cleanup. **Notes**: Python-based, depends on openpyxl. ### [GongRzhe/Office-PowerPoint-MCP-Server](https://github.com/GongRzhe/Office-PowerPoint-MCP-Server) ⭐⭐⭐ (⚠️ archived 2025-12; use anthropics/skills pptx) | Field | Value | |---|---| | License | MIT | | Rating | ⭐⭐⭐ (⚠️ archived) | **What it does**: PPT manipulation via python-pptx — create decks, edit slides, insert images, change layouts. **Audience**: people who want LLMs to auto-generate decks from outlines / Markdown (consultants, lecturers, students). **Notes**: overlaps with `anthropics/skills`'s pptx skill; use this when the official one isn't enough. ### [1weiho/open-slide](https://github.com/1weiho/open-slide) ⭐⭐⭐⭐ | Field | Value | |---|---| | License | MIT | | Rating | ⭐⭐⭐⭐ (agent-native slide framework) | **What it does**: a React slide framework built for coding agents — describe a deck in natural language and let Claude Code / Codex / Cursor write the React; ships two Claude Code Skills (`/create-slide`, `/slide-authoring`). **Audience**: people who want agents to produce decks as code (git-versionable) — a different route from PowerPoint-MCP's .pptx output. **Notes**: TypeScript / React / Vite; scaffold with `npx @open-slide/cli init`. It's an agent-native tool (agents author with it), not a Stage 4 agent-building / orchestration framework. ### [SylphxAI/pdf-reader-mcp](https://github.com/SylphxAI/pdf-reader-mcp) ⭐⭐⭐⭐ | Field | Value | |---|---| | License | MIT | | Rating | ⭐⭐⭐⭐ (high-throughput PDF) | **What it does**: high-speed PDF parsing MCP, ~5-10× faster than `anthropics/skills`'s pdf skill (per their claim). **Audience**: people doing batch reads of papers / contracts / reports. **Notes**: parallel processing; noticeable on large PDFs. ### [tfriedel/claude-office-skills](https://github.com/tfriedel/claude-office-skills) ⭐⭐⭐ | Field | Value | |---|---| | License | NOASSERTION | | Rating | ⭐⭐⭐ (Office skill add-on) | **What it does**: extends `anthropics/skills` with Office workflows it doesn't cover (automation, advanced formatting). **Audience**: people who find the official docx/xlsx/pptx skills too coarse-grained. **Notes**: complements `anthropics/skills`, not a replacement. ### [xberg-io/xberg](https://github.com/xberg-io/xberg) ⭐⭐⭐⭐ | Field | Value | |---|---| | License | NOASSERTION | | Rating | ⭐⭐⭐⭐ | **What it does**: Rust framework for parsing PDF, Office, image, and other common document formats. Provides an MCP server, REST API, and CLI. **Audience**: cross-format batch parsing engineers who care about throughput. **Notes**: covers obscure formats like HWP, ODT, etc., not just PDF / Office. ---

## 3. Google Workspace

Expand selected entries for category 3

### [Google Workspace MCP](https://developers.google.com/workspace/guides/configure-mcp-servers) ⭐⭐⭐⭐ | Field | Value | |---|---| | Delivery | Google hosted remote MCP; OAuth 2.0 | | Rating | ⭐⭐⭐⭐ (**official**, **Developer Preview**) | **What it does**: Google provides dedicated remote MCP servers for Gmail, Drive, Docs, Sheets, Slides, Calendar, Chat, and People. **Audience**: people who want an official entry point for reading Workspace data or creating drafts, updating documents, and scheduling meetings. **Notes**: this is a **Developer Preview**. OAuth 2.0 and user / organization governance apply; enable only the APIs and scopes the task needs. ### [taylorwilsdon/google_workspace_mcp](https://github.com/taylorwilsdon/google_workspace_mcp) ⭐⭐⭐⭐⭐ | Field | Value | |---|---| | License | MIT | | Rating | ⭐⭐⭐⭐⭐ (community-maintained broad Workspace MCP) | **What it does**: Gmail, Calendar, Docs, Sheets, Slides, Drive, Chat, Forms, Tasks, Search — all in one MCP server. **Audience**: people who need community features beyond the official Preview and are willing to maintain the OAuth setup. **Notes**: community-maintained; broad coverage also means a broader permission surface. Start with read-only access and the smallest useful scope. ### [xing5/mcp-google-sheets](https://github.com/xing5/mcp-google-sheets) ⭐⭐⭐⭐ | Field | Value | |---|---| | License | MIT | | Rating | ⭐⭐⭐⭐ (Sheets-only) | **What it does**: focused Google Sheets / Drive integration — create sheets, edit cells, query formulas. **Audience**: people using only Google Sheets who don't want the full Workspace MCP. **Notes**: narrower scope than `google_workspace_mcp`, but simpler setup. ---

## 4. Microsoft 365

Expand selected entries for category 4

### [Softeria/ms-365-mcp-server](https://github.com/Softeria/ms-365-mcp-server) ⭐⭐⭐⭐ | Field | Value | |---|---| | License | MIT | | Rating | ⭐⭐⭐⭐ (full M365) | **What it does**: M365 + Office services via Microsoft Graph API — Outlook, Teams, OneDrive, SharePoint. **Audience**: enterprise M365 users wanting LLM-driven email replies, calendar lookups, OneDrive operations. **Notes**: requires Azure AD app registration; corporate IT policies may block this. ### [ryaker/outlook-mcp](https://github.com/ryaker/outlook-mcp) ⭐⭐⭐ | Field | Value | |---|---| | License | NOASSERTION | | Rating | ⭐⭐⭐ (Outlook only) | **What it does**: Outlook mail / calendar via Graph API. **Audience**: people who only need Outlook, not the rest of M365. **Notes**: narrower scope than `ms-365-mcp-server`. ### [merill/lokka](https://github.com/merill/lokka) ⭐⭐⭐ | Field | Value | |---|---| | License | MIT | | Rating | ⭐⭐⭐ | **What it does**: M365 + Microsoft Graph admin operations — Entra (AD), Intune, etc. **Audience**: M365 system admins managing tenants / users / policies. **Notes**: more useful for IT admins than end users. ---

## 5. Dev Collaboration (GitHub / Atlassian / Slack…)

Expand selected entries for category 5

### [github/github-mcp-server](https://github.com/github/github-mcp-server) ⭐⭐⭐⭐⭐ | Field | Value | |---|---| | License | MIT | | Rating | ⭐⭐⭐⭐⭐ (**official**) | **What it does**: GitHub's official MCP — issues / PRs / repos / Actions / Codespaces. **Audience**: GitHub users who need to look up repositories, organize issues, or assist with PR review. **Notes**: prefer OAuth or a least-privilege token. Keep human approval before creating an issue, changing a PR, or triggering a workflow. Track A exercise CLI-9 in A3 uses this entry point. ### [sooperset/mcp-atlassian](https://github.com/sooperset/mcp-atlassian) ⭐⭐⭐⭐⭐ | Field | Value | |---|---| | License | MIT | | Rating | ⭐⭐⭐⭐⭐ (community-maintained Atlassian MCP) | **What it does**: Connects Confluence and Jira through a community MCP server you can self-host. **Audience**: teams that need self-hosting, custom authentication, or a deployment shape the official remote MCP does not support. **Notes**: community maintained. Choose this or the official Atlassian remote MCP below after checking company IT policy and permission requirements. ### [Atlassian Rovo MCP](https://support.atlassian.com/atlassian-rovo-mcp-server/docs/getting-started-with-the-atlassian-remote-mcp-server/) ⭐⭐⭐⭐⭐ | Field | Value | |---|---| | Delivery | Atlassian hosted remote MCP; OAuth 2.1 | | Rating | ⭐⭐⭐⭐⭐ (**Atlassian official**) | **What it does**: Lets an MCP-capable client read and write Jira, Confluence, and Bitbucket Cloud according to user permissions. **Audience**: teams using Atlassian Cloud. **Notes**: OAuth 2.1 applies; review approval details before changing issues, pages, or repositories. ### [Slack MCP Server](https://docs.slack.dev/ai/mcp-overview/) ⭐⭐⭐⭐⭐ | Field | Value | |---|---| | Delivery | Slack official MCP Server | | Rating | ⭐⭐⭐⭐⭐ (**Slack official**) | **What it does**: Lets an AI app search Slack channels, send messages, manage canvases, and perform other Slack actions. **Audience**: teams using Slack with an MCP-capable client. **Notes**: this is not read-only; confirm organization policy and keep human approval before sending or changing content. ### [Linear MCP](https://linear.app/docs/mcp) ⭐⭐⭐⭐⭐ | Field | Value | |---|---| | Delivery | Linear hosted remote MCP; Streamable HTTP | | Rating | ⭐⭐⭐⭐⭐ (**Linear official**) | **What it does**: Queries or updates Linear issues, projects, and comments. **Audience**: Linear users who manage sprints or backlogs. **Notes**: use the official read-only option when needed; writes follow signed-in user permissions. ### [SaseQ/discord-mcp](https://github.com/SaseQ/discord-mcp) ⭐⭐⭐ | Field | Value | |---|---| | License | MIT | | Rating | ⭐⭐⭐ | **What it does**: Discord MCP — read/write channel messages, manage servers. **Audience**: maintainers running OSS / community Discord servers. **Notes**: requires Discord bot token; watch rate limits. ### [Graphify-Labs/graphify](https://github.com/Graphify-Labs/graphify) ⭐⭐⭐⭐⭐ | Field | Value | |---|---| | License | Apache-2.0 | | Rating | ⭐⭐⭐⭐⭐ | **What it does**: AI coding skill that turns codebases / SQL schemas / R scripts / shell scripts / docs / papers / images / videos into a queryable knowledge graph. Works across Claude Code, Codex, OpenCode, Cursor, Gemini CLI. **Audience**: engineers / researchers analyzing large codebases, tracking cross-file references, or asking questions across "app code + DB schema + infra" together. **Notes**: cross-cutting tool — fits both dev collaboration (understanding existing codebases) and research workflow (turning any artifact into a graph). When stuck on a big codebase, use graphify to extract structure, then feed it back to Claude for reasoning. ### [upstash/context7](https://github.com/upstash/context7) ⭐⭐⭐⭐⭐ | Field | Value | |---|---| | License | MIT | | Rating | ⭐⭐⭐⭐⭐ (coding context) | **What it does**: Finds documentation snippets by library and version, then puts them into the Agent's current context to reduce the chance of using an old API. **Audience**: developers who write code across library versions and want to find relevant documentation before editing. **Notes**: still check the version and original official documentation. It helps find material; it does not guarantee that every result is current or complete. ### [DeusData/codebase-memory-mcp](https://github.com/DeusData/codebase-memory-mcp) ⭐⭐⭐⭐⭐ | Field | Value | |---|---| | License | MIT | | Rating | ⭐⭐⭐⭐⭐ (code intelligence) | **What it does**: Indexes a codebase into a queryable knowledge graph, so a coding agent can inspect structure, symbols, and call paths before returning to the actual code to verify them. **Audience**: people running coding agents on large or unfamiliar repos who want fast orientation and lower token use. **Notes**: re-index after big edits, since the graph can go stale; treat its answers as a fast first pass and verify load-bearing claims (who-calls-X / is-this-dead) against the actual code. ---

## 6. Databases

Expand selected entries for category 6

### [googleapis/mcp-toolbox](https://github.com/googleapis/mcp-toolbox) ⭐⭐⭐⭐⭐ | Field | Value | |---|---| | License | Apache-2.0 | | Rating | ⭐⭐⭐⭐⭐ (**Google official**, multi-DB) | **What it does**: cross-DB MCP server — MySQL / PostgreSQL / Cloud SQL / Spanner / BigQuery. **Audience**: engineers running databases on Google Cloud, or anyone needing multi-engine support. **Notes**: open-source + Google-maintained; solid choice for production use. ### [bytebase/dbhub](https://github.com/bytebase/dbhub) ⭐⭐⭐⭐⭐ | Field | Value | |---|---| | License | MIT | | Rating | ⭐⭐⭐⭐⭐ (community multi-DB) | **What it does**: zero-dependency, token-efficient multi-DB MCP — Postgres, MySQL, SQL Server, MariaDB, SQLite. **Audience**: engineers who don't want the Google Cloud SDK and need cross-OSS-DB support. **Notes**: overlaps with `googleapis/mcp-toolbox` but lighter weight. ### [supabase/mcp](https://github.com/supabase/mcp) ⭐⭐⭐⭐⭐ | Field | Value | |---|---| | License | Apache-2.0 | | Rating | ⭐⭐⭐⭐⭐ (**Supabase official-community**) | **What it does**: connect Supabase (Postgres, Auth, Storage, Edge Functions) to LLMs. **Audience**: full-stack devs using Supabase as backend. **Notes**: official community-maintained. ### [timescale/pg-aiguide](https://github.com/timescale/pg-aiguide) ⭐⭐⭐⭐ | Field | Value | |---|---| | License | Apache-2.0 | | Rating | ⭐⭐⭐⭐ (Postgres coding aid) | **What it does**: MCP server + Claude plugin to help LLMs write better PostgreSQL code. **Audience**: Postgres-heavy SQL writers / DBAs. **Notes**: focused on "LLM writes better SQL", not just query execution. ### [benborla/mcp-server-mysql](https://github.com/benborla/mcp-server-mysql) ⭐⭐⭐⭐ | Field | Value | |---|---| | License | MIT | | Rating | ⭐⭐⭐⭐ (read-only MySQL) | **What it does**: read-only MySQL MCP — let the LLM see schemas, run queries. **Audience**: scenarios where the LLM should analyze production DBs but never modify them. **Notes**: read-only is a safety feature, not a limitation. ### [mongodb-js/mongodb-mcp-server](https://github.com/mongodb-js/mongodb-mcp-server) ⭐⭐⭐⭐ | Field | Value | |---|---| | License | Apache-2.0 | | Rating | ⭐⭐⭐⭐ (**MongoDB official**) | **What it does**: MongoDB and MongoDB Atlas Cluster MCP server. **Audience**: engineers using MongoDB / Atlas. **Notes**: `mongodb-js` is MongoDB's official GitHub org. ### [redis/mcp-redis](https://github.com/redis/mcp-redis) ⭐⭐⭐⭐ | Field | Value | |---|---| | License | MIT | | Rating | ⭐⭐⭐⭐ (**Redis official**) | **What it does**: official Redis MCP — natural-language operations on Redis and Redis Stack (Vector / Search / JSON). **Audience**: people using Redis as cache / vector DB / queue. **Notes**: officially maintained; includes vector search. ### [awslabs/mcp](https://github.com/awslabs/mcp) ⭐⭐⭐⭐ | Field | Value | |---|---| | License | Apache-2.0 | | Rating | ⭐⭐⭐⭐ (**AWS official**) | **What it does**: AWS's first-party MCP servers (Lambda / S3 / DynamoDB / CloudWatch / Cost Explorer and more). **Audience**: teams on AWS who want agents to query / operate their cloud. **Notes**: officially maintained by AWS; uses your existing AWS login (CLI profiles / IAM roles), no separate token to manage. ---

## 7. Browser Automation / Web Scraping

Expand selected entries for category 7

### [microsoft/playwright-mcp](https://github.com/microsoft/playwright-mcp) ⭐⭐⭐⭐⭐ | Field | Value | |---|---| | License | Apache-2.0 | | Rating | ⭐⭐⭐⭐⭐ (**Microsoft official**) | **What it does**: Playwright MCP server — let the LLM open browsers, click buttons, fill forms, scrape pages. **Audience**: anyone doing E2E automation, cross-site integration, scraping behind logins. **Notes**: Playwright is an official project; evaluate it alongside the browser permissions and login flows your task needs. ### [ChromeDevTools/chrome-devtools-mcp](https://github.com/ChromeDevTools/chrome-devtools-mcp) ⭐⭐⭐⭐⭐ | Field | Value | |---|---| | License | Apache-2.0 | | Rating | ⭐⭐⭐⭐⭐ (**Chrome official**) | **What it does**: expose Chrome DevTools to coding agents — performance, network, console traces all available to the LLM. **Audience**: developers debugging frontend bugs, doing web performance analysis. **Notes**: pairs perfectly with Playwright MCP — one drives, one observes. ### [firecrawl/firecrawl-mcp-server](https://github.com/firecrawl/firecrawl-mcp-server) ⭐⭐⭐⭐⭐ | Field | Value | |---|---| | License | MIT | | Rating | ⭐⭐⭐⭐⭐ (**Firecrawl official**) | **What it does**: Firecrawl's official MCP — large-scale web scraping + search + structured extraction. **Audience**: people scraping large amounts of web data for training / RAG / research. **Notes**: requires Firecrawl API key (has a free tier). ### [browserbase/mcp-server-browserbase](https://github.com/browserbase/mcp-server-browserbase) ⭐⭐⭐⭐ (⚠️ archived) | Field | Value | |---|---| | License | Apache-2.0 | | Rating | ⭐⭐⭐⭐ (**Browserbase official**, ⚠️ archived) | **What it does**: Browserbase's official MCP, paired with Stagehand for cloud-based browser automation. **Audience**: people whose local browser automation is too heavy / who need parallel cloud sessions. **Notes**: commercial service (free tier exists); complementary to Playwright MCP (local vs cloud). ---

## 8. Design (Figma / Excalidraw)

Expand selected entries for category 8

### [Canva MCP](https://www.canva.dev/docs/mcp/) ⭐⭐⭐⭐⭐ | Field | Value | |---|---| | Delivery | Canva hosted remote MCP; `https://mcp.canva.com/mcp` | | Rating | ⭐⭐⭐⭐⭐ (**Canva official**) | **What it does**: Lets an AI assistant create, edit, search, and export Canva designs, and work with assets, brands, and comments. **Audience**: people who want to operate Canva while following existing design permissions. **Notes**: each user authenticates; operations follow asset and design permissions, and some features vary by plan. Review high-impact edits before approving them. ### [GLips/Figma-Context-MCP](https://github.com/GLips/Figma-Context-MCP) ⭐⭐⭐⭐⭐ | Field | Value | |---|---| | License | MIT | | Rating | ⭐⭐⭐⭐⭐ (community-maintained design-to-code MCP) | **What it does**: feed Figma layout info to coding agents — read design files, expose component structure, let Cursor / Claude Code generate matching React components. **Audience**: front-end devs going from Figma designs to component code. **Notes**: requires a Figma access token. Start with a test file and least privilege while checking whether it fits your design-to-code workflow. ### [excalidraw/excalidraw-mcp](https://github.com/excalidraw/excalidraw-mcp) ⭐⭐⭐⭐⭐ | Field | Value | |---|---| | License | NOASSERTION | | Rating | ⭐⭐⭐⭐⭐ (**Excalidraw official**) | **What it does**: streamable Excalidraw MCP — let LLMs draw architecture diagrams and flowcharts directly. **Audience**: anyone writing design docs / system architecture / flowcharts who wants Claude to draw from text. **Notes**: official Excalidraw; output imports straight into Excalidraw for editing. ### [yctimlin/mcp_excalidraw](https://github.com/yctimlin/mcp_excalidraw) ⭐⭐⭐⭐ | Field | Value | |---|---| | License | MIT | | Rating | ⭐⭐⭐⭐ (alternative Excalidraw) | **What it does**: MCP server + Claude Code Skill, real-time canvas sync, create / edit / export. **Audience**: people who need real-time canvas sync and programmatic operation. **Notes**: complementary to the official; community-maintained. ### [pbakaus/impeccable](https://github.com/pbakaus/impeccable) ⭐⭐⭐⭐⭐ | Field | Value | |---|---| | License | Apache-2.0 | | Rating | ⭐⭐⭐⭐⭐ | **What it does**: "**The design language that makes your AI harness better at design.**" A vocabulary / pattern set that helps AI agents produce UI / visual output that escapes the generic "AI-generated" feel. **Audience**: developers using AI to generate UI / mockups / visual designs but getting generic results; front-end + AI workflows. **Notes**: not an MCP server or Skill bundle — it's a **design language** reference. Feed AI the higher-quality design vocabulary and it produces better output. ---

## 9. Monitoring / Observability

Expand selected entries for category 9

### [grafana/mcp-grafana](https://github.com/grafana/mcp-grafana) ⭐⭐⭐⭐⭐ | Field | Value | |---|---| | License | Apache-2.0 | | Rating | ⭐⭐⭐⭐⭐ (**Grafana official**) | **What it does**: Grafana's official MCP — query dashboards / metrics / alerts from the LLM. **Audience**: SREs / DevOps using Grafana for metrics. **Notes**: "why did this dashboard line drop?" — ask, and the LLM pulls metrics for the answer. ### [getsentry/sentry-mcp](https://github.com/getsentry/sentry-mcp) ⭐⭐⭐⭐ | Field | Value | |---|---| | License | NOASSERTION | | Rating | ⭐⭐⭐⭐ (**Sentry official**) | **What it does**: query Sentry error events / issues / traces from LLMs. **Audience**: engineers using Sentry for production errors. **Notes**: "show me last week's stack trace for this error" works directly in Claude Code. ### [winor30/mcp-server-datadog](https://github.com/winor30/mcp-server-datadog) ⭐⭐⭐ | Field | Value | |---|---| | License | Apache-2.0 | | Rating | ⭐⭐⭐ (community Datadog) | **What it does**: Datadog API MCP — monitors / logs / metrics. **Audience**: Datadog users while there's no official Datadog MCP yet. **Notes**: likely to be replaced once Datadog ships an official MCP. ---

## 10. Media / Streaming (YouTube / Spotify)

Expand selected entries for category 10

### [varunneal/spotify-mcp](https://github.com/varunneal/spotify-mcp) ⭐⭐⭐⭐ | Field | Value | |---|---| | License | MIT | | Rating | ⭐⭐⭐⭐ | **What it does**: connect LLMs to Spotify — play tracks, manage playlists, query history. **Audience**: anyone integrating playback control or text → music workflows with Claude Code. **Notes**: requires Spotify Premium (API restriction). ### [kimtaeyoon83/mcp-server-youtube-transcript](https://github.com/kimtaeyoon83/mcp-server-youtube-transcript) ⭐⭐⭐⭐ | Field | Value | |---|---| | License | MIT | | Rating | ⭐⭐⭐⭐ (YouTube transcripts) | **What it does**: pull YouTube video transcripts into the LLM for summary / translation / RAG. **Audience**: people using video as study material, batch-summarizing YouTube content. **Notes**: depends on YouTube auto-captions; non-English transcripts are hit-or-miss. ### [ZubeidHendricks/youtube-mcp-server](https://github.com/ZubeidHendricks/youtube-mcp-server) ⭐⭐⭐⭐ | Field | Value | |---|---| | License | MIT | | Rating | ⭐⭐⭐⭐ (full YouTube API) | **What it does**: full YouTube API MCP — beyond transcripts, also video management, Shorts, analytics. **Audience**: YouTube creators automating channel management. **Notes**: requires YouTube Data API key + OAuth. ---

## 11. Chinese-language Ecosystem

Expand selected entries for category 11

### [leemysw/feishu-docx](https://github.com/leemysw/feishu-docx) ⭐⭐⭐ | Field | Value | |---|---| | License | MIT | | Rating | ⭐⭐⭐ | **What it does**: bidirectional Feishu (Lark) docs / sheet / bitable ↔ Markdown, with OAuth 2.0, CLI, TUI, Claude Skills. **Audience**: Chinese-language users on Feishu / Lark wanting to bridge Lark content with Claude Code. **Notes**: community maintained. Lark APIs, OAuth scopes, and supported features can change; check the current documentation and start in a test space. ### [netease-youdao/LobsterAI](https://github.com/netease-youdao/LobsterAI) ⭐⭐⭐⭐ | Field | Value | |---|---| | License | MIT | | Rating | ⭐⭐⭐⭐ | **What it does**: NetEase Youdao's product-style Agent demonstrates workflow automation, cross-app coordination, and file processing. **Audience**: people evaluating a Chinese interface and integrations with mainland Chinese services. **Notes**: this is a complete Agent product, not a Skill or MCP server. Check the current project documentation for supported integrations, permissions, and deployment options. ### [QwenLM/Qwen-Agent](https://github.com/QwenLM/Qwen-Agent) ⭐⭐⭐⭐⭐ | Field | Value | |---|---| | License | Apache-2.0 | | Rating | ⭐⭐⭐⭐⭐ | **What it does**: Alibaba's official Qwen agent framework — RAG, tool use, code interpreter, multi-agent, MCP-compatible. Defaults to Qwen models but swappable to other LLMs. **Audience**: developers using Qwen / Tongyi as primary LLM; teams that want a Chinese-native agent framework (examples + docs are bilingual but Chinese-first). **Notes**: MCP compatibility and replaceable models are the main teaching points; check current releases, examples, and supported hosts before adopting it. ### [coze-dev/coze-studio](https://github.com/coze-dev/coze-studio) ⭐⭐⭐⭐⭐ | Field | Value | |---|---| | License | Apache-2.0 | | Rating | ⭐⭐⭐⭐⭐ | **What it does**: open-source release of ByteDance Coze — no-code agent builder (workflow / plugin / knowledge / memory), self-hosted or cloud. **Audience**: teams building agents without writing code; engineers wanting a reference implementation of an enterprise agent platform (RAG, workflow, memory, plugin system). **Notes**: built on Coze's in-house Eino framework; connects to OpenAI / Claude / Qwen / domestic Chinese LLMs. Powers both the international (coze.com) and mainland (coze.cn) products. ### [coze-dev/coze-loop](https://github.com/coze-dev/coze-loop) ⭐⭐⭐⭐ | Field | Value | |---|---| | License | Apache-2.0 | | Rating | ⭐⭐⭐⭐ | **What it does**: Coze's agent observability + evaluation platform — trace, debug, eval, prompt management. The back half of the agent dev lifecycle. **Audience**: teams whose agents are running in production and need monitoring; developers wanting to see how "agent eval / observability" can be designed. **Notes**: peer to LangSmith / Arize Phoenix; OSS release is self-hostable. ### [liaokongVFX/LangChain-Chinese-Getting-Started-Guide](https://github.com/liaokongVFX/LangChain-Chinese-Getting-Started-Guide) ⭐⭐⭐⭐ | Field | Value | |---|---| | License | unspecified | | Rating | ⭐⭐⭐⭐ | **What it does**: Chinese-language LangChain getting-started guide — covers basics, prompts, memory, agents, chains, and applied examples. It is a structured Chinese learning resource. **Audience**: Chinese-language users who want LangChain but find the English docs heavy; readers who want to understand LangChain's design before committing to the framework. **Notes**: no formal license (content is openly readable); LangChain itself moves fast — some APIs in the guide may diverge from the latest version. ### [chatchat-space/Langchain-Chatchat](https://github.com/chatchat-space/Langchain-Chatchat) ⭐⭐⭐⭐ | Field | Value | |---|---| | License | Apache-2.0 | | Rating | ⭐⭐⭐⭐ | **What it does**: LangChain-based open-source knowledge-base QA system — local deployment, supports multiple vector stores, end-to-end RAG example. **Audience**: Chinese teams who want RAG without building it from scratch; scenarios requiring local-only deployment (no cloud LLM). **Notes**: useful for reading the end-to-end structure of a locally deployed RAG system. Maintenance has slowed; before using it for a new project, check the current branch, dependencies, and issues, and treat it primarily as a reference implementation. ### [usewhale/whale](https://github.com/usewhale/whale) ⭐⭐⭐ | Field | Value | |---|---| | License | MIT | | Rating | ⭐⭐⭐ | **What it does**: Terminal AI coding assistant optimized for DeepSeek models — supports MCP server integration, Claude-style Skills, conversation caching, written in Go. **Audience**: Chinese developers who use DeepSeek as their primary LLM; those who want a terminal tool without the full Claude Code stack. **Notes**: One of the few open-source tools with DeepSeek-specific optimization; MCP + Skills dual support allows incremental capability expansion. ### [simonlin1212/a-stock-data](https://github.com/simonlin1212/a-stock-data) ⭐⭐⭐⭐ | Field | Value | |---|---| | License | Apache-2.0 | | Rating | ⭐⭐⭐⭐ | **What it does**: China A-share data Skill that wraps mootdx, EastMoney, akshare, iwencai, and other sources as a data entry point an AI coding assistant can call. **Audience**: Chinese developers using Claude Code / Codex / OpenClaw for investment research or quantitative analysis; those who don't want to build data-fetching logic from scratch. **Notes**: community implementation. Data-source terms, stability, and fields can change; verify the original data and licenses before investment research. Check the project's current documentation for compatible hosts. > Looking for WeChat / DingTalk integrations? Today the mainstream is chatbot frameworks (e.g., zhayujie/CowAgent), not pure MCP servers. Will add when proper MCPs emerge. ### [MoonshotAI/Kimi-K2](https://github.com/MoonshotAI/Kimi-K2) ⭐⭐⭐ | Field | Value | |---|---| | License | Modified MIT | | Rating | ⭐⭐⭐ | **What it does**: Moonshot's Kimi K2 open-weight LLM series — open weights + OpenAI/Anthropic-compatible API, oriented toward agentic / coding / long-horizon tasks; usable as a backend model for an agent stack. **Audience**: Chinese developers who want to run agent / coding workflows on a domestic open model, or to self-host open weights. **Notes**: License is Modified MIT (standard MIT + added large-scale-commercial clauses) — read the original LICENSE before commercial use; weights are also available on Hugging Face. ### [zai-org/GLM-4.5](https://github.com/zai-org/GLM-4.5) ⭐⭐⭐ | Field | Value | |---|---| | License | Apache-2.0 | | Rating | ⭐⭐⭐ | **What it does**: Zhipu (Z.ai)'s GLM-4.5 open model — positioned as Agentic, Reasoning, and Coding (ARC) foundation models; open weights + API, usable as a backend for agent / tool use / coding. **Audience**: Chinese developers evaluating domestic open agentic models, or who need weights under a permissive license (Apache-2.0). **Notes**: zai-org is Zhipu's open-source org; the same series also has GLM-4 () for context; weights are on Hugging Face. ---

## 12. Other Common (Cloudflare / Stripe…)

Expand selected entries for category 12

### [cloudflare/mcp-server-cloudflare](https://github.com/cloudflare/mcp-server-cloudflare) ⭐⭐⭐⭐⭐ | Field | Value | |---|---| | License | Apache-2.0 | | Rating | ⭐⭐⭐⭐⭐ (**Cloudflare official**) | **What it does**: Cloudflare's official MCP — Workers, Pages, R2, KV, D1, DNS, Zero Trust. **Audience**: anyone running edge / serverless on Cloudflare. **Notes**: officially maintained; compare its permissions and supported operations with the edge workflows you need. ### [stripe/ai](https://github.com/stripe/ai) ⭐⭐⭐⭐ | Field | Value | |---|---| | License | MIT | | Rating | ⭐⭐⭐⭐ (**Stripe official**) | **What it does**: Stripe's official AI agent toolkit, includes an MCP server — handle payments, subscriptions, refunds, customers. **Audience**: developers wiring payment / billing into agent flows. **Notes**: ⚠️ this is real money. Test thoroughly in sandbox before going to production. ### YIELD INTELLIGENCE MCP (Hosted Remote Server) | Field | Value | |---|---| | Type | hosted MCP server | | Rating | ⭐⭐⭐ (finance analysis tool; practical example of hosted vs self-hosted MCP architecture) | **What it does**: Uses a hosted remote MCP to read US Treasury yields and income-asset analysis, while showing the connection shape of hosted versus self-hosted MCP. **Audience**: learners who want to observe MCP requests and results with test finance questions; it is not an automatic trading entry point. **Notes**: the [live endpoint](https://api.intuitek.ai/yield/mcp) is provided by the MIT [thebrierfox/intuitek-ace](https://github.com/thebrierfox/intuitek-ace) repository. Data timing, methods, authentication, and pricing can change; check current service documentation and original market data before use. Not investment advice. ### [ComposioHQ/composio](https://github.com/ComposioHQ/composio) ⭐⭐⭐⭐⭐ | Field | Value | |---|---| | License | MIT | | Rating | ⭐⭐⭐⭐⭐ (many tool-integration hub) | **What it does**: a platform (SDKs + MCP servers) that connects agents to many apps (Slack / GitHub / Gmail / Salesforce / Notion…) and handles the logins for you, so you don't build a separate connector for each one. **Audience**: teams whose agents need broad API coverage without maintaining dozens of separate MCP servers. **Notes**: provides MCP servers + Python / TypeScript SDKs; connect to Claude Code via MCP. A "tool aggregator" (compare with n8n / Zapier for automation). --- ### [morluto/jacobian](https://github.com/morluto/jacobian) ⭐⭐⭐ (⚠️ submitted by its author) | Field | Value | |---|---| | License | MIT | | Rating | ⭐⭐⭐ | **What it teaches**: a directly installable math MCP server for practicing agent calls to composable, exact computations across polynomial maps, linear algebra, and graph algorithms. **Who it's for**: researchers and developers who want to add mathematical computation to MCP workflows or have agents work with structured mathematical problems. **Notes**: a Python project, but the launcher ships on npm, which is why it starts with `npx`. Provides an MCP server, CLI, and Python library. Start with a local MCP configuration and use the native Python API when needed. **How to run**: ```bash npx -y jacobian mcp ``` ---

## 13. Research Workflow Skills (academic / paper / lit) Research Skills can organize literature and writing workflows; the researcher remains responsible for citations, source data, and academic integrity.

Expand selected entries for category 13

> ⚠️ **Maintainer's own projects**: these are research tools the repo maintainer [@WenyuChiou](https://github.com/WenyuChiou) uses and has published. Inclusion is based on the workflow problem solved, not popularity; evaluate them against your research requirements, data, and host. ### [WenyuChiou/ai-research-skills](https://github.com/WenyuChiou/ai-research-skills) ⭐⭐⭐⭐⭐ | Field | Value | |---|---| | License | MIT | | Rating | ⭐⭐⭐⭐⭐ (full research workflow) | **What it does**: Uses a set of research Skills to cover literature triage, research design, project context, manuscript writing, and multi-AI delegation, distributed through a marketplace. **Audience**: grad students / postdocs wanting a complete "research workflow" skill set in one drop. **Notes**: marketplace format, aligns with the plugin/marketplace concept taught in Stage 5.4. ### [WenyuChiou/academic-writing-skills](https://github.com/WenyuChiou/academic-writing-skills) ⭐⭐⭐ | Field | Value | |---|---| | License | MIT | | Rating | ⭐⭐⭐ (narrow but deep) | **What it does**: rigorous academic paper writing / revision / submission skill for Claude Code. Field-agnostic, customizable per-paper via journal_format.md and style_overrides.md. **Audience**: researchers actively writing / revising papers who want to automate banned-word audit, figure-text coupling, submission checklists. **Notes**: use it through the complete marketplace, or install it separately by following the current official instructions. ### [WenyuChiou/zotero-skills](https://github.com/WenyuChiou/zotero-skills) ⭐⭐⭐⭐ | Field | Value | |---|---| | License | MIT | | Rating | ⭐⭐⭐⭐ | **What it does**: Zotero CLI skill — programmatically search, add, classify, annotate references. **Audience**: Zotero users wanting Claude Code to organize their library directly. **Notes**: complementary to [`MuiseDestiny/zotero-gpt`](https://github.com/MuiseDestiny/zotero-gpt) — that one is a Zotero plugin (chat inside Zotero), this one is a CLI / Skill (operate Zotero from Claude Code). ### [WenyuChiou/research-hub](https://github.com/WenyuChiou/research-hub) ⭐⭐⭐⭐ | Field | Value | |---|---| | License | MIT | | Rating | ⭐⭐⭐⭐ | **What it does**: AI-operable research workspace integrating Zotero, Obsidian, and Gemini Notebook (formerly NotebookLM), with several interfaces. **Audience**: researchers using Zotero, Obsidian, and Gemini Notebook (formerly NotebookLM) together. **Notes**: complementary to single-tool MCPs (mcp-obsidian, notion-mcp, etc.) — this is a hub that integrates multiple tools. ---

## 14. Multi-LLM Delegation Skills Delegation tools need clear files, budgets, acceptance checks, and stop conditions. A second model does not automatically make an answer correct.

Expand selected entries for category 14

> ⚠️ **Maintainer's own projects**: these delegation skills come from the maintainer's daily workflow. Inclusion depends on whether each entry can bound responsibility, inputs, outputs, and acceptance—not popularity. Multi-LLM tools change quickly; evaluate them with the production framework in Stage 7. ### How delegation skills compose The skills below can work together, but each one needs a clear task boundary and acceptance check: ![Claude + 3 delegate skills — division of labor](../resources/diagrams/multi-llm-delegation-composition.en.png) Do not turn model names into permanent job titles. First check the current models, tools, and costs; then assign **design / review, implementation, and long-form synthesis** to the suitable executors. Use the same acceptance criteria for the final result. ### [WenyuChiou/codex-delegate](https://github.com/WenyuChiou/codex-delegate) ⭐⭐⭐⭐⭐ | Field | Value | |---|---| | License | MIT | | Rating | ⭐⭐⭐⭐⭐ | **What it does**: Claude Code skill that uses Codex CLI as the execution specialist — multi-file refactors, batch edits, boilerplate generation, wrapper-based implementation tasks. Claude writes the plan + reviews; Codex executes. **Audience**: developers wanting to save tokens / accelerate large-scale mechanical edits; learners who want to verify "multi-agent isn't just a buzzword". **Use it for**: applying one transform across many files, generating test scaffolds, porting an existing pattern, or writing migration scripts. **Don't use for**: tasks without a clear responsibility boundary or acceptance criteria, or tasks that still need an independent security review. **Notes**: treat it as a bounded executor, not a permanent job assigned to one model. Validate the result with the same acceptance criteria. ### [WenyuChiou/gemini-delegate-skill](https://github.com/WenyuChiou/gemini-delegate-skill) ⭐⭐⭐ (⚠️ archived 2026-07) | Field | Value | |---|---| | License | MIT | | Rating | ⭐⭐⭐ (⚠️ archived) | **What it does**: Historical Claude Code delegation skill that sent long-form, cross-document, or CJK tasks to another CLI executor. **Audience**: maintainers of older setups and readers studying early multi-LLM delegation patterns. **Use it for**: history and migration reference only. **Don't use for**: new work; the repository is archived and is not a current installation starting point. **Notes**: do not preserve its old fixed model roles. Use a maintained delegation entry that fits the current host and can be checked against explicit acceptance criteria. ### [WenyuChiou/agent-collab-skills](https://github.com/WenyuChiou/agent-collab-skills) ⭐⭐ | Field | Value | |---|---| | License | MIT | | Rating | ⭐⭐ (experimental — treat as reference) | **What it does**: Claude Code marketplace for multi-agent collaboration — task splitter, output reconciler, adversarial debate, shared memory, acceptance gate. **Audience**: people coordinating multiple delegates who want to see one way of packaging multi-agent coordination into a marketplace. **Notes**: **experimental** — don't treat this as a framework ready for production use. It's the maintainer's own setup made public as a reference. For multi-agent frameworks built for production, see LangGraph / Microsoft Agent Framework / CrewAI in Stage 7. ---

## 15. Finance / Trading Agents > ⚠️ **Application-domain section**: agents applied to quantitative trading, hedge-fund simulation, and automated order placement. The two entries here are Apache-2.0 and MIT, but licensing across this category varies widely — verify each repo before reuse. **Caveat**: real-money trading agents carry significant risk; listed here for agent-design study, not as investment advice.

Expand selected entries for category 15

### [TauricResearch/TradingAgents](https://github.com/TauricResearch/TradingAgents) ⭐⭐⭐ | Field | Value | |---|---| | License | Apache-2.0 | | Rating | ⭐⭐⭐ | **What it does**: a multi-agent LLM framework for financial trading decisions, with bull / bear / fundamentals / technicals / risk agents collaborating. **Audience**: learners studying how multi-agent systems split analytical work; quant researchers experimenting with LLM augmentation of existing pipelines. **Notes**: Apache-2.0 — modification and commercial use permitted (retain license notice). **Not investment advice — do not run on real funds directly.** ### [virattt/ai-hedge-fund](https://github.com/virattt/ai-hedge-fund) ⭐⭐⭐ | Field | Value | |---|---| | License | MIT | | Rating | ⭐⭐⭐ | **What it does**: a multi-role AI hedge-fund simulation where bull / bear / fundamentals / technicals / risk agents collaborate to produce trade recommendations. **Audience**: Stage 7 multi-agent learners wanting a complete application example; people interested in the agent × finance crossover. **Notes**: MIT-licensed; same caveat as above. **Simulation only — not investment advice.** ---

## 16. Web Search / Retrieval

Expand selected entries for category 16

### [exa-labs/exa-mcp-server](https://github.com/exa-labs/exa-mcp-server) ⭐⭐⭐⭐ | Field | Value | |---|---| | License | MIT | | Rating | ⭐⭐⭐⭐ (**Exa official**) | **What it does**: Exa's official MCP — LLM / agent-oriented web search (neural + keyword) that returns clean results to feed straight into prompts. **Audience**: people doing research / fact-check / online RAG retrieval — semantic search shines for "concept-related" queries, less so for pure keyword lookups. **Notes**: requires an Exa API key. ### [tavily-ai/tavily-mcp](https://github.com/tavily-ai/tavily-mcp) ⭐⭐⭐⭐ | Field | Value | |---|---| | License | MIT | | Rating | ⭐⭐⭐⭐ (LLM search MCP) | **What it does**: wraps the Tavily search API as MCP, giving an Agent search results and sources. **Audience**: beginners who want an Agent to search the web and then verify the source content separately. **Notes**: requires a Tavily API key. Plans and quotas can change; check the current official documentation. ---

## 17. Security / MCP Governance

Expand selected entries for category 17

### [trailofbits/skills](https://github.com/trailofbits/skills) ⭐⭐⭐⭐⭐ | Field | Value | |---|---| | License | CC-BY-SA-4.0 | | Rating | ⭐⭐⭐⭐⭐ (**official Trail of Bits**) | **What it does**: the official Claude Code plugin marketplace from security firm Trail of Bits, carrying the skills they use themselves for security analysis, testing, and vulnerability research. **Audience**: people who want an agent to help with code audits, vulnerability analysis, and security testing; also anyone curious how a professional security team encodes its own methodology as skills. **Notes**: install with `/plugin marketplace add trailofbits/skills`; Codex supports the same marketplace directly. The license is **CC-BY-SA-4.0** (share-alike), so derivative work ships under the same terms; check that before commercial integration. Note this is a **different repo** from [`trailofbits/skills-curated`](https://github.com/trailofbits/skills-curated) in the Stage 5 table: that one is a curation marketplace of community-contributed plugins reviewed by Trail of Bits staff, this one is Trail of Bits' own security skills. ### [stacklok/toolhive](https://github.com/stacklok/toolhive) ⭐⭐⭐⭐ | Field | Value | |---|---| | License | Apache-2.0 | | Rating | ⭐⭐⭐⭐ | **What it does**: runs every MCP server in an isolated container, replaces local credentials with a minimal permission file, and adds audit logs plus configurable identity and access policy. Ships a Kubernetes operator as well. **Audience**: teams where everyone has installed their own pile of MCP servers and somebody now has to answer "who installed what, and which credentials can it reach". **Notes**: Go project, available as a desktop app and a CLI. Open-source ToolHive and Stacklok's enterprise product are separate, so check which features sit on which side. ### [NVIDIA/OpenShell](https://github.com/NVIDIA/OpenShell) ⭐⭐⭐⭐ | Field | Value | |---|---| | License | Apache-2.0 | | Rating | ⭐⭐⭐⭐ (**NVIDIA official**) | **What it does**: a sandboxed execution environment for autonomous agents, with declarative policy over files, network and processes. Claude Code and Codex run inside it unmodified; OpenClaw goes through NemoClaw, the next entry below. **Audience**: people who want an agent to actually execute commands without handing it the whole machine. **Notes**: Rust project. Needs Docker / Podman or host virtualization for MicroVM-backed sandboxes; Windows goes through WSL 2 and is marked experimental. **⚠️ The project still calls itself alpha**, so read it and try it, but assess before depending on it. The split against `toolhive`: toolhive governs MCP servers, this governs the Agent's own execution boundary. ### [nolabs-ai/nono](https://github.com/nolabs-ai/nono) ⭐⭐⭐⭐ | Field | Value | |---|---| | License | Apache-2.0 | | Rating | ⭐⭐⭐⭐ | **What it does**: confines agents using the operating system's own isolation (Linux Landlock, macOS Seatbelt), with no daemon and no container. **Audience**: people who want a sandbox without installing a whole container runtime for it. **Notes**: Rust project, maintained by the team behind Sigstore. Going container-free is the main trade against OpenShell: faster to start, but the isolation is only as strong as what the OS provides. ## What's not here? If your integration isn't above, check these catalogs first: - [Official MCP Registry](https://registry.modelcontextprotocol.io/) — find published MCP servers, then check the maintainer, permissions, license, and current status - [`wong2/awesome-mcp-servers`](https://github.com/wong2/awesome-mcp-servers) — community MCP server catalog with many entries by category - [`punkpeye/awesome-mcp-servers`](https://github.com/punkpeye/awesome-mcp-servers) — another MCP server catalog, complementary - [`modelcontextprotocol/servers`](https://github.com/modelcontextprotocol/servers) — official reference implementations for learning how MCP works; not production recommendation material. - [`travisvn/awesome-claude-skills`](https://github.com/travisvn/awesome-claude-skills) — Claude Skills catalog ### Want to add something? 1. Open an issue with the repo link, why it should be added, and which category it fits. 2. Or open a PR with its source, license or service status, editorial rating, purpose, audience, permissions, and limits. 3. Explain which learning gap it fills; popularity numbers alone do not prove quality or safety. Read [`resources/style-guide.en.md`](style-guide.en.md) and [`CONTRIBUTING.en.md`](../CONTRIBUTING.en.md) before submitting. --- ## Notes for anyone helping out later Not an SLA — just "do what you can" guidance: - Treat vendor documentation or the canonical repo as the source for official status, licensing, permissions, and hosted endpoints. - CI periodically scans repository redirects, archive status, HTTP errors, and freshness signals. A warning still needs human judgment; do not remove a stable tool only because it has not released recently. - Define the reader's task and safety boundary before opening a new category, then add verifiable official or high-quality community entries. - Apply the same learning-value, maintenance, licensing, and permission standards to Chinese-community tools; do not use a star threshold. - Make wording and formatting clear enough for a five-year-old to follow while preserving exact terms, limits, and sources.