# Gen-AI-Experiments: a course repository of 130 plus notebooks, agent skills and MCP servers > This repository is a teaching collection for generative AI work, holding more than 130 runnable applications, a set of installable SKILL.md agent skills, Model Context Protocol servers and clients, and day-one cookbooks for the newest models. It is organised as a path from starter to advanced projects, and it is a body of examples rather than a library. **buildfastwithai/gen-ai-experiments** — Collection of Jupyter notebooks is designed to provide you with a comprehensive guide to various AI tools and technologies - Repository: https://github.com/buildfastwithai/gen-ai-experiments - Website: https://www.buildfastwithai.com/genai-course - Stars: 788 · Forks: 224 - Language: Jupyter Notebook - License: MIT - Published: 2026-10-20 · Updated: 2026-10-20 · Language: en - Canonical page: https://hysenlabs.com/projects/buildfastwithai-gen-ai-experiments ## What the collection covers The repository describes itself as a curated collection of more than 130 production-ready generative AI apps, agents, installable agent skills, MCP servers and latest-model cookbooks, built with Claude, GPT-5, Gemini 3, Qwen, GLM, the Model Context Protocol, LangChain, RAG and multi-agent teams. The project description is shorter, calling it a collection of Jupyter notebooks designed to provide a comprehensive guide to various AI tools and technologies. The top-level entries show how the content is partitioned. There are cookbooks, skills, mcp, apps, agents, ai-creations, browser-plugins, workflows and an archive directory, plus assets and a CONTRIBUTING file. The archive directory is worth noticing, because it is where content that has been retired usually goes, and its presence suggests the collection is maintained rather than accumulated. The stated aims are learn by building with 130 or more applications you can run, modify and learn from, a structured learning path with starter, intermediate and advanced projects, tutorials across more than 100 libraries including LangChain, LlamaIndex, CrewAI, AG2 and Weaviate, multi-language support across English, Hindi and other Indian languages, and hands-on cookbooks for the newest models from Anthropic, OpenAI, Google, Qwen, GLM, Kimi and DeepSeek. The project is MIT licensed, the default branch is main, the last push was on 2026-09-22, and there are no GitHub releases, so the content moves by commit rather than by version. ## Agent skills as self contained markdown The Agent Skills part is the most distinctive thing in the repository, and it works by being almost nothing but documentation. An Agent Skill here is a self-contained SKILL.md file, with optional scripts, references and assets directories, that teaches an agent a repeatable workflow. Drop one into Claude Code, Cowork or Codex and it becomes a capability rather than a piece of advice. The README is explicit that these are model-agnostic and reusable, and the skills index is linked for full details. Three skills are listed in detail. The Landing Page Generator produces high-converting landing pages as production HTML using PAS, AIDA and BAB copy frameworks, with call to action strategy, search engine meta and conversion and speed audits. The Talking Avatar builds realtime voice chat applications with a lip-synced character avatar on the OpenAI Realtime API, using Vite or Next.js with your own key. The Crazy Ecommerce Builder turns a brand brief into an art-directed storefront with original generated product photography. Then there are three skill packs. The frontend pack holds a Frontend Core document and a Tailwind Component Factory, plus style systems named Glass UI, Neo-Brutalism, Minimal Luxury, Bold SaaS, Editorial and Retro-Futurist, each as its own file. The backend pack holds an MCP Server Builder for scaffolding production servers, a Next.js Route Handler, MERN auth best practices and a Mongoose Schema Architect. The docs and research pack holds a README Architect, a Research Synthesizer and a Deck Outline Generator. The structure is the point. A skill is a prompt with a contract, and the pack layout means a team can adopt one style system and one backend pattern without taking the rest. ## The Model Context Protocol part The MCP section explains the protocol as the open standard for connecting agents to tools and data, and then states that the repository ships servers, clients, integrations and runnable examples under the mcp directory. The directories are specified for contributors: servers go in mcp/servers/, clients in mcp/clients/, integrations in mcp/integrations/, and runnable demos in mcp/examples/. Four entries are named. Launch MCP is a zero-dependency MCP server that turns any GitHub repository into a complete launch kit, producing repo analysis, release notes, a changelog, a blog post, posts for X, Reddit, Hacker News and LinkedIn, an interactive HTML demo and an SVG share card, with the tools analyze_repo and generate_share_card. The MCP-Use App under apps/mcp-use is a reference client showing how to wire MCP tools into an agent loop. The MCP Workshop is a guided hands-on notebook for building a first MCP server and client, and an AG2 cookbook notebook demonstrates a multi-agent GroupChat with tool use over MCP. The choice of an MCP-Use reference client next to a workshop notebook is deliberate and worth noting. A protocol is easy to describe and hard to get right, and having both a minimal working client and a guided build exercise is a better way to learn it than a specification. ## Model cookbooks organised by provider The cookbooks are described as day-one testing notebooks for the newest frontier and open models, organised by provider under the cookbooks/models directory. The Claude section shows the pattern. There are cookbooks for Claude Opus 4.8, Opus 4.7 and Opus 4.6 Fast, then Sonnet 4.6, Sonnet 4.5 and Haiku 4.5, each as a separate notebook. The file names alternate between a hyphenated model name and an underscore separated, capitalised form, which reflects that the notebooks were added over time rather than renamed as a set. That per-model, per-version layout is the single most useful fact about the cookbooks. A model cookbook is only interesting for a few weeks after a release, because what changes is the behaviour of the model, not the technique. Organising by provider and by exact version means a reader can tell at a glance how old a test result is, and the maintainers can add a new file without touching the old ones. The same day-one approach appears in the wider model list, which names Claude Opus 4.8, GPT-5.5, Gemini 3, Qwen3.6, GLM-5, Kimi K2.7 and DeepSeek V4, along with the note that these are hands-on cookbooks for the newest models. ## How to read the repository as a learning path The README lists a course site at buildfastwithai.com and a genai-course page, so the repository is the material behind a course rather than a standalone reference. Given the size, the directory layout is the practical way in. An agent building a landing page would start in skills rather than cookbooks, an engineer adding tools to an agent would start in mcp, and someone evaluating whether a new model is worth switching to would start in cookbooks/models. Each of those directories has its own convention, and the skills and MCP sections both state where contributed work should go. The multi-language support is an unusual feature for a repository of this kind. The projects are described as supporting English, Hindi and other Indian languages, which means the examples are written with localisation in mind rather than being English-only code that a translation layer is bolted onto. For a repository whose stated audience includes developers in India, that is a design decision rather than an afterthought. The library coverage claim is the other thing to weigh. More than 100 libraries are covered, which sounds like breadth but is also a maintenance question, since each library's API changes and a notebook that used to run may not run any more. The presence of an archive directory and a CONTRIBUTING guide suggests the collection is curated, but a notebook in the archive should be read as a historical artefact rather than a working example. ## Where Gen-AI-Experiments is the wrong choice The first limitation is that none of this is a library. There is no package to install, no API to depend on and no stability promise, so a team that wants a maintained component should look elsewhere. What is here is a body of examples, and examples have a different contract: they show a shape, they do not carry it. The second is the shelf life of the model cookbooks. They are explicitly day-one tests for models whose behaviour changes with each release, so a notebook that proved something about Opus 4.8 tells you little about the model you would use next year, and might not even run against it. The per-version organisation makes that visible, which is honest, but it means the repository is a moving target rather than a reference you can cite. The third is the breadth versus depth trade. A repository that covers more than 100 libraries and more than 130 applications cannot go deep on any of them, and the practical consequence is that each example is a starting point you will rewrite. For a learner that is exactly right. For an engineer trying to settle a production question about a specific library, a focused book or the library's own documentation will be more useful. The skills are the part that stands on its own, because a skill is a workflow description rather than a demonstration. A SKILL.md that teaches an agent to scaffold an MCP server or audit a landing page can be installed and used as is, and that makes the skills directory the part of this repository with a life longer than the notebooks. ## Conclusion Gen-AI-Experiments fits a developer learning generative AI by building, who wants a working example of a skill, an MCP server or a model cookbook to copy rather than read about. It does not fit a team that needs a maintained library with a stable API, and the model cookbooks are by definition dated the day they are written. Before drawing on it, check which part you actually need, since the skills, the MCP servers and the notebooks live in separate directories with different conventions, and read the model cookbooks as snapshots: the repository itself is organised by provider and model version, which is a good sign that the authors know those files expire. ## FAQ ### What is in the Gen-AI-Experiments repository? It is a curated collection of more than 130 production-ready generative AI applications and experiments, installable agent skills, Model Context Protocol servers, clients and integrations, and cookbooks for the newest models, organised into cookbooks, skills, mcp, apps, agents, ai-creations, browser-plugins, workflows and an archive directory. The project is MIT licensed. ### What is an Agent Skill and how do I use one? A skill is a self-contained SKILL.md file, with optional scripts, references and assets, that teaches an agent a repeatable workflow and is model-agnostic. Drop one into Claude Code, Cowork or Codex and it becomes a capability. The repository ships skills for landing page generation, talking avatars, ecommerce building, plus frontend, backend and docs skill packs. ### What does the Model Context Protocol section of the repository contain? It contains MCP servers, clients, integrations and runnable examples under the mcp directory, with a convention that servers go in mcp/servers/, clients in mcp/clients/, integrations in mcp/integrations/ and demos in mcp/examples/. Named entries include the Launch MCP server, the MCP-Use reference client, a hands-on MCP workshop notebook and an AG2 multi-agent GroupChat notebook using MCP tools. ### How are the model cookbooks organised? As day-one testing notebooks under cookbooks/models/, grouped by provider, with one notebook per model and version, such as Claude Opus 4.8, Opus 4.7, Opus 4.6 Fast, Sonnet 4.6, Sonnet 4.5 and Haiku 4.5, alongside cookbooks for the other named frontier and open models. ### Is this repository suitable as a production dependency? No. It is a teaching collection of examples rather than a library, with no package to install and no API stability promise, and the model cookbooks are dated snapshots of behaviour that changes with each model release. The skills are the part most usable as-is, because a skill is a workflow description rather than a demonstration. ## Sources - [buildfastwithai/gen-ai-experiments on GitHub](https://github.com/buildfastwithai/gen-ai-experiments) - [Issues](https://github.com/buildfastwithai/gen-ai-experiments/issues) - [License: MIT](https://github.com/buildfastwithai/gen-ai-experiments/blob/main/LICENSE) - [Project website](https://www.buildfastwithai.com/genai-course) - [README](https://github.com/buildfastwithai/gen-ai-experiments/blob/main/README.md) --- Hysen Labs editorial analysis, written from the project's own repository and release notes. Cite the canonical page: https://hysenlabs.com/projects/buildfastwithai-gen-ai-experiments