Skill Finder
The best Claude skills for building AI agents
Start with Google Agents CLI from Google if you are building agents: at 89 it has the top score here without a high-risk scan warning, has 533,836 installs, and walks Claude from spec to deploy with Google's ADK. skill-creator scores 91 for writing Claude skills, but at least one scanner rated it high risk. The list also covers mcp-builder, ai-sdk, Mastra, Genkit, LangGraph and model training guides.
The 12 skills made for this job
-
1
Get Google Agents CLIGoogle Agents CLI
For developers building AI agents with Google's ADK who want guidance from scaffold to deploy. Skip it if your agents will not run on Google Cloud.
-
2
Get skill-creatorskill-creator
For anyone writing their own Claude skills who wants test runs and a review loop. Skip it if you only install skills that other people made.
-
3 Get mcp-builder
mcp-builder
For developers turning an API or service into tools Claude can call through MCP. Skip it if you only want to use a server someone else already built.
-
4
Get ai-sdkai-sdk
For TypeScript developers adding chat, agents or structured output with Vercel's AI SDK. Skip it if your app is in Python or calls a model provider's own SDK.
-
5
Get mastramastra
For TypeScript developers building agents and workflows with Mastra who want Claude to check current APIs first. Skip it if you do not use Mastra.
-
6
Get mcp-apps-buildermcp-apps-builder
For TypeScript developers building MCP servers or interactive MCP Apps with the mcp-use framework. Skip it if you write MCP servers in Python or with a different SDK.
-
7
Get deep-agents-memorydeep-agents-memory
For developers building LangChain Deep Agents that need to remember things between threads. Skip it if you use a different agent framework or only need one-off chats.
-
8
Get developing-genkit-jsdeveloping-genkit-js
For TypeScript developers building AI flows or agents with Genkit who want Claude to use current APIs. Skip it if you work in another language or AI framework.
-
9 Get langgraph-fundamentals
langgraph-fundamentals
For developers building agent workflows in LangGraph with Python or TypeScript. Skip it if you are not using LangGraph or a single prompt or simple chain already does the job.
-
10
Get healthhealth
For Claude Code, Codex or Pi users whose agent ignores instructions or whose hooks and MCP servers misbehave. Skip it if you are chasing an app bug or want a pull request review.
-
11
Get AI Research SkillsAI Research Skills
For ML engineers who want working code patterns for fine-tuning, serving and evaluating models. Skip it if you only call hosted model APIs and never train or host models.
-
12
Get Context Engineering SkillsContext Engineering Skills
For engineers building agent systems who want written guidance on context, memory and evaluation. Skip it if you want a tool that runs tasks rather than a set of explanations.
Also useful for agents
Their main job is something else, but they help with this one.
-
Get Azure skillsAzure skills
For teams on Microsoft Azure who want Claude to plan, check and run deployments or build Foundry agents. Skip it if you do not use Azure.
-
Get herdrherdr
For developers who run several coding agents in Herdr panes and want Claude to steer them. Skip it if you do not use the Herdr terminal multiplexer.
-
Get Cloudflare skillsCloudflare skills
For developers building on Cloudflare who want help picking products and working from current docs. Skip it if your app runs on a different platform.
-
Get gemini-api-devgemini-api-dev
For developers writing Python or TypeScript code against the Gemini API and its managed agents. Skip it if you build on another model provider or never call Google's API.
-
Get stitch-loopstitch-loop
For builders who want an agent to grow a multi-page site with Google Stitch, one page per run. Skip it if you lack the Stitch MCP server or want to design pages yourself.
-
Get tavily-best-practicestavily-best-practices
For developers adding Tavily web search, extraction or crawling to their own agent, RAG system or app. Skip it if you just want Claude itself to search the web for you.
-
Get book-to-skillbook-to-skill
For readers who want a technical book's frameworks on hand as a skill Claude can consult. Skip it if a plain summary of the book would do.
-
Get prompt-masterprompt-master
For anyone who wants a rough idea turned into a tight prompt for a specific AI tool. Skip it if you rarely write prompts or want lessons in prompting theory.
-
Get Claude Skills (alirezarezvani)Claude Skills (alirezarezvani)
For teams who want one large library of role-based skills to pick from. Skip it if you want a few focused skills you can read through in an afternoon.
How to choose
Google Agents CLI agrees a spec with you one question at a time, scaffolds with agents-cli, then evaluates, deploys and publishes. It needs the agents-cli tool, uv, and a Google Cloud project for deploys. Skip it if your agents will not run on Google Cloud.
mcp-builder, from Anthropic, writes an MCP server that gives Claude tools for an outside API. It plans the endpoints, builds with the TypeScript SDK or Python FastMCP, checks the result with MCP Inspector and writes ten test questions. It needs Node.js or Python and access to that API.
For TypeScript, pick by framework. ai-sdk writes code with Vercel's ai package and needs a model provider or AI Gateway key. mastra and developing-genkit-js make Claude read current docs for Mastra or Genkit before coding, and mcp-apps-builder builds MCP servers with the mcp-use framework.
skill-creator drafts a SKILL.md, runs test prompts with and without the skill in subagents, and shows pass rates in a browser viewer. It needs Python and the claude CLI. At least one scanner rated it high risk, so read its scan results before you install it.
LangChain users get langgraph-fundamentals and deep-agents-memory, and health audits your own agent setup. AI Research Skills, 98 guides to training and serving models, and Context Engineering Skills, 24 guides on what reaches an agent's context, each carry a high-risk rating from at least one scanner.