🍪 A chatbot is a recipe book that talks back. An agent is the cook who also raids the fridge.
We serve the agent world in bite-size, source-linked snacks.
New to AI agents? Grab a cookie. Six tiny steps, plain English, about 10 minutes. Steps 1–3 need zero coding. Here to make images or videos? Jump to [I make stuff 🎨](#i-make-stuff-the-creator-track). ## 1. What's an AI agent? A **chatbot** talks. An **agent** talks *and then gets up and does stuff*: it opens files, uses tools and keeps going, step by step, until the job's done. Anthropic's engineers put it like this: agents "dynamically direct their own processes and tool usage" ([Anthropic: Building effective agents](https://www.anthropic.com/engineering/building-effective-agents)). ## 2. Chatbot or agent: which do you need? - **Asking, drafting, explaining?** A chat app has you covered: [ChatGPT](https://chatgpt.com/) or [Claude](https://claude.ai/). - **Work across files, code or other apps?** That's agent territory. Agents need permission to act, so they also need a bit more care (see step 5). ## 3. Pick your first tool by goal | Your goal | Start with | Coding needed? | |---|---|---| | Write, summarize, brainstorm | [ChatGPT](https://chatgpt.com/) or [Claude](https://claude.ai/) | Nope | | Build or fix code in an editor | [Cursor](https://cursor.com/) or [GitHub Copilot](https://github.com/features/copilot) | No, but you can | | Let an agent work in your terminal | [Claude Code](https://code.claude.com/docs/en/overview) | No, but you can | Claude Code runs in a terminal. You chat, it codes. Pick **one** and stick with it for a week. Tasting every flavor on day one just gives you a stomach ache. 🍪 ## 4. Your first 15-minute task 1. Pick something small you can check, like "turn these meeting notes into 5 bullets" or, in a code tool, "explain what this file does." 2. Set the table: say who it's for, how long, and what "good" looks like. 3. Read the answer, then ask: "What did you assume?" (It's a great question to ask anyone.) 4. Taste before you serve: check the result yourself before you use it anywhere. ## 5. Safety basics - **Keep secrets off the plate.** No passwords, API keys or private customer data in prompts. - **Double-check the answer.** AI can sound very sure and still be wrong. Check facts, numbers and code. - **Say yes on purpose.** Tools like Cursor ask before they run outside tools, by default ([Cursor docs](https://cursor.com/docs/mcp)). Read what it wants to do before you click. - **Only add extras from people you trust.** Anthropic and Cursor both say this about skills and MCP servers, because add-ons can run code ([Anthropic: Agent Skills](https://platform.claude.com/docs/en/agents-and-tools/agent-skills/overview), [Cursor docs](https://cursor.com/docs/mcp)). ## 6. Glossary (the jargon jar) - **MCP (Model Context Protocol):** a shared plug that lets AI apps connect to other stuff like files, databases and tools. The official docs call it "a USB-C port for AI applications" ([modelcontextprotocol.io](https://modelcontextprotocol.io/docs/getting-started/intro)). One plug, many gadgets. - **Skill:** a recipe card for your agent: a folder of instructions (a `SKILL.md` file, plus optional scripts) for a job it does often. It only comes out of the drawer when it's needed ([Anthropic: Agent Skills](https://platform.claude.com/docs/en/agents-and-tools/agent-skills/overview)). - **Plugin / connector:** a ready-made snack pack for one app. In ChatGPT, plugins bundle skills, MCP servers and optional screens ([OpenAI plugins](https://developers.openai.com/plugins)). - **Context window:** everything the AI can "see" at once while it answers, its short-term memory. A bigger plate isn't always a better meal ([Anthropic docs](https://platform.claude.com/docs/en/build-with-claude/context-windows)). ## 🎨 I make stuff: the creator track Agents can help run a creator channel now: make the pictures, cut the video, add the voice, and get it ready to publish. Same agent idea as coding — just different snacks on the plate. Four bites: 1. **Pick one maker tool.** [Runway's plugin](https://github.com/runwayml/runway-mcp-plugin) makes images, video and audio. [Pika](https://github.com/Pika-Labs/pika-create-plugin) and [MiniMax MCP](https://github.com/MiniMax-AI/MiniMax-MCP) do the creative rounds too. More on the [Creators shelf](/shelf/#creators) — samples first so you can taste before you plug in. 2. **Plug it into your agent.** In Cursor, type `/add-plugin` and search for Runway. You sign in with your Runway account, no API key, and it spends your plan's credits ([Runway plugin](https://github.com/runwayml/runway-mcp-plugin)). 3. **Order a small snack first.** One image. Then one short clip. Then a line of voiceover. Like it? Ask for tweaks in plain words. 4. **Make it yours — then ship it.** Anthropic's [canvas-design skill](https://github.com/anthropics/skills/tree/main/skills/canvas-design) tells Claude to make original designs and never copy existing artists. Good manners for humans too. When the snack looks right, publish. 📺 Watch: [Introducing Runway MCP](https://www.youtube.com/watch?v=KhXxcU6cgzM) (0:45, Runway) **Next:** take this week's [First Bite](/archive/001/#first-bite) in Snack Pack #1, or [subscribe](/subscribe/) for one short email a week. ## Video Tasting (official tutorials) ### Getting started with Claude.ai (5:19, beginner) https://www.youtube.com/watch?v=0vZ_UVLhSQQ Channel: Anthropic · Published 2025-12-02 After watching you can: Have your first good chat with Claude: ask well, add files, and pick a model or style. - A friendly tour of the Claude chat screen, plus prompting basics. - How to add context and files, tweak replies (styles, model choice) and use research mode. - Wraps up with tips for working with Claude over the long haul. - Watch just 0:56 Prompting basics: https://www.youtube.com/watch?v=0vZ_UVLhSQQ&t=56s - Watch just 2:05 Adding context and files: https://www.youtube.com/watch?v=0vZ_UVLhSQQ&t=125s (Summary source: chapters + description) ### Getting started with projects in Claude.ai (7:08, beginner) https://www.youtube.com/watch?v=GJ5jTgcbRHA Channel: Anthropic · Published 2025-12-02 After watching you can: Set up a Claude Project once, so you stop re-explaining yourself every chat. - What Projects are: a lunchbox that keeps your instructions and files across chats. - Make your first project, write its instructions and add knowledge. - Share it and set permissions for your team. - Watch just 1:33 Creating your first project: https://www.youtube.com/watch?v=GJ5jTgcbRHA&t=93s - Watch just 2:21 Defining project instructions: https://www.youtube.com/watch?v=GJ5jTgcbRHA&t=141s (Summary source: chapters + description) ### Getting started with connectors in Claude.ai (3:43, beginner) https://www.youtube.com/watch?v=_jjSS0qGFbI Channel: Anthropic · Published 2025-12-11 After watching you can: Hook Claude up to the apps and files you already use, and keep those hookups tidy. - Connectors let Claude reach your files, apps and workflows. - How to set up new connectors and manage the ones you have. - Ends with a real-world workflow you can copy. - Watch just 0:50 Setting up new connectors: https://www.youtube.com/watch?v=_jjSS0qGFbI&t=50s - Watch just 2:25 Real-world workflow example: https://www.youtube.com/watch?v=_jjSS0qGFbI&t=145s (Summary source: chapters + description) ### Claude Agent Skills Explained (3:14, intermediate) https://www.youtube.com/watch?v=fOxC44g8vig Channel: Anthropic · Published 2025-11-26 After watching you can: Tell skills, CLAUDE.md, MCP servers and subagents apart, and know when to use each. - Skills are folders of expertise Claude loads automatically when a task matches. - Only each skill's name and description load at startup; the full SKILL.md loads on demand. - Compares skills with CLAUDE.md (project facts), MCP (connects to data) and subagents (fixed roles). - Watch just 0:30 How Agent Skills work: https://www.youtube.com/watch?v=fOxC44g8vig&t=30s - Watch just 1:42 Agent Skills vs MCP Servers: https://www.youtube.com/watch?v=fOxC44g8vig&t=102s - Watch just 2:33 Stacking skills into one bigger agent: https://www.youtube.com/watch?v=fOxC44g8vig&t=153s (Summary source: transcript + chapters + visual analysis) ### Why we built—and donated—the Model Context Protocol (MCP) (35:32, intermediate) https://www.youtube.com/watch?v=PLyCki2K0Lg Channel: Anthropic · Published 2025-12-11 After watching you can: Explain what MCP is, why it exists, and why it's now an open, foundation-run standard. - Stuart Ritchie talks with MCP co-creator David Soria Parra about where the protocol came from. - Covers the problem MCP solves, the USB-C analogy, community adoption and criticisms. - Explains why Anthropic donated MCP to the Linux Foundation. - Watch just 0:00 What is MCP?: https://www.youtube.com/watch?v=PLyCki2K0Lg&t=0s - Watch just 2:46 The USB-C analogy: https://www.youtube.com/watch?v=PLyCki2K0Lg&t=166s - Watch just 32:53 Advice for non-developers: https://www.youtube.com/watch?v=PLyCki2K0Lg&t=1973s (Summary source: chapters + description) ### Getting started with research in Claude.ai (2:50, beginner) https://www.youtube.com/watch?v=R-KJgjIrh24 Channel: Anthropic · Published 2025-12-02 After watching you can: Kick off a Research task in Claude and know when it's worth the wait. - How to start a Research task and what it's doing while you sip your coffee. - The jobs it's best at: digging across many sources. - How it teams up with extended thinking. - Watch just 0:00 Initiating research: https://www.youtube.com/watch?v=R-KJgjIrh24&t=0s - Watch just 1:38 Ideal use cases: https://www.youtube.com/watch?v=R-KJgjIrh24&t=98s (Summary source: chapters + description) ### Let Claude handle work in your browser (1:35, beginner) https://www.youtube.com/watch?v=rBJnWMD0Pho Channel: Anthropic · Published 2025-12-18 After watching you can: See a browser agent in action: Claude for Chrome looking, clicking and typing for you. - Claude for Chrome is a browser extension: Claude can see, click, type and move around web pages. - Three demos: dashboard data into one doc, fixing slide comments, and testing a Claude Code build in Chrome. - Basically 'chatbot vs agent', live. (Summary source: description) ### Introducing Cowork: Claude Code for the rest of your work (1:09, beginner) https://www.youtube.com/watch?v=UAmKyyZ-b9E Channel: Anthropic · Published 2026-01-12 After watching you can: Get Cowork: hand Claude a big task, point it at your stuff, come back to a finished file. - Cowork takes your time-eating tasks and hands back finished work. - It pulls from your local files, cloud tools and the web. - Out come spreadsheets, slides, documents and PDFs. (Summary source: description) ### Your tools are now interactive in Claude (0:54, beginner) https://www.youtube.com/watch?v=bluAmTHoEow Channel: Anthropic · Published 2026-01-26 After watching you can: See what 'interactive tools inside Claude' actually looks like. - Your connected tools now work right inside the Claude chat. - Examples: Asana projects, Slack drafts, Amplitude charts, Figma diagrams. - No tab-hopping required. (Summary source: description) ### Introducing Runway MCP (0:45, beginner) https://www.youtube.com/watch?v=KhXxcU6cgzM Channel: Runway · Published 2026-05-27 After watching you can: Picture how an agent makes images and videos for you through Runway MCP. - Runway MCP connects Runway to Claude, ChatGPT, Cursor, Replit and more. - Make images and videos with models like Gen-4.5, Seedance 2.0, GPT Images 2.0 and Kling. - All from the tool you already work in. (Summary source: description) ### Using Workflows | Runway Academy (2:35, beginner) https://www.youtube.com/watch?v=MsVl_CKu_gk Channel: Runway · Published 2025-11-21 After watching you can: Run a featured Runway workflow and tweak it into your own little tool. - Workflows let you build your own custom tools inside Runway. - Start from a featured workflow, run it, then see how its steps connect. - Customize it and make it yours. - Watch just 0:25 Using featured workflows: https://www.youtube.com/watch?v=MsVl_CKu_gk&t=25s - Watch just 0:44 Running a workflow: https://www.youtube.com/watch?v=MsVl_CKu_gk&t=44s - Watch just 1:27 Customizing workflows: https://www.youtube.com/watch?v=MsVl_CKu_gk&t=87s (Summary source: chapters + description) ## Creator corner: making images and video with agents - [Plugin] ElevenLabs plugin (2026-10-05): ElevenLabs' official plugin gives your agent a voice. Skills for text-to-speech, music, sound effects, dubbing and more, plus their hosted MCP server. Sign in with your ElevenLabs account; no API key needed for the hosted server. Source: https://github.com/elevenlabs/plugin · Sample media: https://elevenlabs.io/docs/api-reference/voices/search (ElevenLabs: official preview of the premade voice "Roger") - [Skill] Anthropic canvas-design skill (2026-04-20): A recipe card that teaches Claude to make posters and art as .png or .pdf. It writes a short design philosophy first, then paints it. Original designs only: the skill says never copy existing artists. Source: https://github.com/anthropics/skills/tree/main/skills/canvas-design · Sample media: https://github.com/anthropics/skills/tree/main/skills/canvas-design (Anthropic: Introducing Agent Skills (canvas-design)) - [MCP server] MiniMax MCP (2026-08-20): MiniMax's official server gives your agent text_to_image and generate_video tools (plus voices). Latest update: moved to MCP SDK v2 and retired the Music tool. Needs a MiniMax API key. Source: https://github.com/MiniMax-AI/MiniMax-MCP · Sample media: https://github.com/MiniMax-AI/MiniMax-MCP#4-generate-images (MiniMax (official README)) - [Plugin] Runway plugin (Runway MCP) (2026-09-11): Runway's official plugin plugs your agent into Runway's hosted MCP server. Ask for images, video, music or speech right in chat. Sign in with your Runway account (no API key); generations spend your plan's credits. Source: https://github.com/runwayml/runway-mcp-plugin · Sample media: https://www.youtube.com/watch?v=KhXxcU6cgzM (Runway (official YouTube): Introducing Runway MCP) - [MCP server] Luma API MCP (2025-04-18): Luma's official server: create images (Photon) and videos (Ray) from your MCP client. Needs a Luma API key and a quick setup script. Older pick: last updated April 2025. Source: https://github.com/lumalabs/luma-api-mcp · Sample media: https://www.youtube.com/watch?v=qfWoP7wNVJY (Luma (official YouTube): Introducing Ray 3.2) - [Skill] Anthropic algorithmic-art skill (2026-04-20): Teaches Claude to make generative art with p5.js: flow fields, particles, seeded randomness. Original art only, per the skill. Source: https://github.com/anthropics/skills/tree/main/skills/algorithmic-art · Sample media: https://github.com/anthropics/skills/tree/main/skills/algorithmic-art (Anthropic: Introducing Agent Skills) - [Skill] OpenAI speech skill (2026-02-02): Turns text into voiceover, narration or audio prompts with OpenAI's built-in voices. One clip or a whole batch. Needs an OPENAI_API_KEY. Source: https://github.com/openai/skills/tree/main/skills/.curated/speech · Sample media: https://platform.openai.com/docs/guides/text-to-speech (OpenAI (official TTS docs): Alloy voice sample) - [Skill] OpenAI imagegen skill (2026-04-15): Codex's built-in recipe card for making and editing images: mockups, sprites, textures, cutouts. By default it uses Codex's own image tool, so no API key. Source: https://github.com/openai/skills/tree/main/skills/.system/imagegen · Sample media: https://platform.openai.com/docs/guides/image-generation (OpenAI (official Images API docs sample)) - [MCP server] Figma MCP server (skills 2.2.157) (2026-10-15): Figma's official server lets agents read your designs and, on the remote server, draw right on the canvas. It can even make placeholder images (it asks first and uses Figma AI credits). The latest bump added figma-spec and figma-code-workflows skills. Source: https://github.com/figma/mcp-server-guide · Sample media: https://www.youtube.com/watch?v=s7BZz03icnY (Figma (official YouTube): What is the Figma MCP server?) - [Plugin] Pika plugin 0.19.3 (2026-10-14): Pika's official plugin lets your agent make and edit images, video, music, sound effects and voice. Comes with task skills for ads, headshots, camera moves, podcasts and more. Uses the credits on your Pika account. Source: https://github.com/Pika-Labs/pika-create-plugin · Sample media: https://www.youtube.com/watch?v=zIyjTj2dk10 (Pika Labs (official YouTube): The new Pika platform) - [Video] Introducing Runway MCP (2026-05-27): Runway MCP connects Runway to Claude, ChatGPT, Cursor, Replit and more. Make images and videos with models like Gen-4.5, Seedance 2.0, GPT Images 2.0 and Kling. All from the tool you already work in. Source: https://www.youtube.com/watch?v=KhXxcU6cgzM - [Video] Using Workflows | Runway Academy (2025-11-21): Workflows let you build your own custom tools inside Runway. Start from a featured workflow, run it, then see how its steps connect. Customize it and make it yours. Source: https://www.youtube.com/watch?v=MsVl_CKu_gk ## Daily snacks (full text) # Agent Snacks — Daily snacks ## 2026-10-15 · Copilot: Copilot eats your CodeQL backlog Assign a CodeQL alert to Copilot: it fixes, re-runs CodeQL to prove the alert is gone, then opens a draft PR. Up to 25 alerts per batch. Now GA. Spice: 2 · Source: https://github.blog/changelog/2026-10-15-agentic-autofix-for-code-scanning-alerts-general-availability ## 2026-10-14 · Claude Code: Claude Code naps smart Leave a 200K+ token chat idle ~54 minutes and Claude Code compacts it before the prompt cache expires, so your next message doesn't re-chew the whole transcript. Opt out with `"idleCompaction": false`. Spice: 2 · Source: https://github.com/anthropics/claude-code/releases/tag/v2.1.299 ## 2026-10-13 · Cline: Cline swapped your default model Never pinned a model? Cline v4.1.24 now defaults Cline + OpenRouter users to Claude Sonnet 5.5. Pin it, or peek at the cost dashboard before blaming the agent. 🍪 Spice: 3 · Source: https://github.com/cline/cline/releases/tag/v4.1.24 ## Snack Pack #1 — full text (https://agentsnacks.ai/archive/001/) # Agent Snacks — Snack Pack #1 > 🍪 Week of Oct 9–16, 2026 · 7 snacks · ~4 min > Agents learned to babysit themselves this week. Grab a coffee — they’ll keep chewing. Hey — welcome to **Snack Pack #1**. Same deal every Thursday: only what *actually shipped* in the AI coding-agent tools you already use, each bite with a source link. No funding fluff. No “AI will change everything.” Just the version bumps that might change your Friday.Nutrition Facts
Serving: 1 Snack Pack · ~8 min
Counts from this issue. “Actually useful” is the editor's call: 5 of 7 change a workflow. Hype: 0 g.
Real workflow shift, but enterprise-policy gated
**What changed:** Assign a CodeQL alert to Copilot and it wanders the repo, proposes a fix, re-runs CodeQL to prove the alert dies, then opens a draft PR. Batch up to **25** alerts into one PR, or poke it via the REST API (`copilot-swe-agent[bot]`). Default-branch alerts only; an enterprise admin has to flip on the “Agentic autofix” policy; it spends Code Security AI Credits + Actions minutes. **Who should care:** Security folks staring at a CodeQL mountain on GHEC / GHAS. **Source:** [GitHub Changelog](https://github.blog/changelog/2026-10-15-agentic-autofix-for-code-scanning-alerts-general-availability) ## 2. HydraFusion stops making you pick a model (and adds a “Max” mode) (Oct 14) 🌶️Research preview; nice, not urgent
**What changed:** Copilot’s multi-model orchestrator (still research preview) refreshed its pool — Claude Opus 5.5, GPT-6 Astra, GPT-6.1 Sol, GPT-6 Luna — drafts on medium effort, reviews on high. Cascade gets a judging step; Critique can tool-snoop the workspace with parallel critics. New **HydraFusion Max** is the “go deep” button. Lives in VS Code, the Copilot app, and Copilot CLI. **Who should care:** Copilot Pro+ / Business folks tired of the model roulette wheel. **Source:** [GitHub Changelog](https://github.blog/changelog/2026-10-14-hydrafusion-updates-and-new-max-option) ## 3. Claude Code: idle chats now compact before the cache expires (v2.1.299–300, Oct 14) 🌶️🌶️Automatic on long sessions; know the opt-out
**What changed:** Leave a 200K+ token terminal chat alone ~54 minutes and it compact-*snacks* itself before the one-hour prompt cache dies — so your next message doesn’t re-chew the whole transcript. Tune with `CLAUDE_CODE_IDLE_COMPACT_MIN_TOKENS` or kill it with `"idleCompaction": false`. Also: `CLAUDE_CODE_DISABLE_FILE_MENTIONS` (paths only, not file guts), and OTel now splits cache writes into `cache_creation_5m_tokens` / `cache_creation_1h_tokens`. **Who should care:** Anyone with marathon Claude Code sessions — or a token bill that makes them wince. **Source:** [v2.1.299](https://github.com/anthropics/claude-code/releases/tag/v2.1.299) · [v2.1.300](https://github.com/anthropics/claude-code/releases/tag/v2.1.300) ## 4. Copilot CLI: hooks that don’t hang the babysitter (v1.0.96–98, Oct 12–14) 🌶️🌶️🌶️Unattended Autopilot loops stop hanging
**What changed:** - **v1.0.96:** `/add-dir` grants sandbox access for the session; interactive settings nudge you to mask likely secrets. - **v1.0.97:** agent-scoped declarative hooks; Autopilot now **denies** a preToolUse “ask” instead of freezing on a dialog nobody is watching. 🫡 - **v1.0.98:** Gemini 4 Argon support; `/ask` expands with Ctrl+E, copies with Ctrl+Y. **Who should care:** Terminal Copilot users running Autopilot / unattended loops. **Source:** [v1.0.96](https://github.com/github/copilot-cli/releases/tag/v1.0.96) · [v1.0.97](https://github.com/github/copilot-cli/releases/tag/v1.0.97) · [v1.0.98](https://github.com/github/copilot-cli/releases/tag/v1.0.98) ## 5. Cursor Origin opens the snack aisle (App Marketplace, Oct 13) 🌶️Matters mostly to tool makers and CI wiring
**What changed:** Origin plugs into 20+ apps — Cloudflare, Vercel, Buildkite, CircleCI, CodeRabbit, Greptile, Semgrep, Socket, Sentry, Linear, and friends. Public submissions welcome. Apps auth with app JWT + install tokens, get webhooks, write check runs / PR comments / reviews. **Who should care:** Dev-tool makers hunting distribution, and teams wiring CI/review bots into Origin. **Source:** [Cursor changelog](https://cursor.com/changelog) ## 6. Cline quietly swapped your default model (v4.1.24, Oct 13) 🌶️🌶️🌶️Unpinned users are silently on a new model
**What changed:** No model pinned? Cline + OpenRouter now default to **Claude Sonnet 5.5**. Catalog refresh adds Haiku 5.5; ~20 other providers also shuffle defaults — **unpinned users will wake up on a different model**. Bonus: turning reasoning off no longer 400s on several frontier models. Desktop v0.0.46 adds a clickable prompt outline. **Who should care:** Cline users who never pinned a model. Check the cost dashboard before you blame the agent. **Source:** [v4.1.24](https://github.com/cline/cline/releases/tag/v4.1.24) · [desktop-v0.0.46](https://github.com/cline/cline/releases/tag/desktop-v0.0.46) ## 7. OpenCode 1.19: the on-ramp to OpenCode 2 (and auto-update is off) (Oct 11) 🌶️🌶️Upgrades are now manual; plan the jump to 2
**What changed:** `opencode upgrade --major` (or `/update` in the TUI) walks you to OpenCode 2. Updates are **no longer automatic** — upgrade on purpose. Also allows OpenAI’s `ultrafast` service tier. **Who should care:** OpenCode folks. Deliberate upgrades > surprise majors. **Source:** [v1.19.0](https://github.com/anomalyco/opencode/releases/tag/v1.19.0) --- ## The Main Course 🍽️ ### MCP vs plugins vs skills: what actually plugs your agent into the world? **Jargon jar:** an *agent* is an AI that does things, not just chats. A *tool* is one thing it can do (like "search issues"). The *context window* is everything it can see at once: its short-term memory, or plate size ([Anthropic docs](https://platform.claude.com/docs/en/build-with-claude/context-windows)). Every tool on the shelf this week claims to "connect your agent." They connect it in three different ways, and picking the wrong one is how you end up with 40 tools in context and an agent that still can't find your deploy script. Here's the menu. **MCP servers = the hands.** The [Model Context Protocol](https://modelcontextprotocol.io/docs/getting-started/intro) is an open standard for connecting AI apps to external systems: data, tools and workflows. The official docs call it a USB-C port for AI, and Claude, ChatGPT, VS Code and Cursor all support it. A server exposes **tools** (functions the model can call), **resources** (data it can read) and **prompts** (templated workflows). In Cursor that means a `stdio` command for local servers, or a Streamable HTTP/SSE URL for remote ones, set in `.cursor/mcp.json` (project) or `~/.cursor/mcp.json` (global) ([Cursor docs](https://cursor.com/docs/mcp)). Write once, plug into any MCP client. That's the whole pitch. *Use MCP when* the agent needs live data or needs to *do* something: open a PR, query a database, read a ticket. Make it a remote HTTP server when a team will share it. **Skills = the recipe card.** An [Agent Skill](https://platform.claude.com/docs/en/agents-and-tools/agent-skills/overview) is a folder with a `SKILL.md`, which has YAML frontmatter (`name`, `description`) plus instructions, and can bundle scripts and reference files. The trick is progressive disclosure. Only the name and description load at startup (~100 tokens per skill). The body (under 5k tokens) loads when a request matches. Bundled files cost nothing until they're read, and scripts run through bash so only their *output* enters context. In Claude Code, skills live in `~/.claude/skills/` (personal) or `.claude/skills/` (project). *Use a skill when* the problem is know-how, not access: "how we write migrations," "our release checklist," "fill this PDF form." Skills teach the agent the moves. They don't give it new hands. **App plugins/connectors = the meal kit.** Plugins bundle the other two for one app. OpenAI's [plugin docs](https://developers.openai.com/plugins) describe ChatGPT plugins as skills + MCP servers + optional UI, packaged and submitted to the ChatGPT Directory, with a guide for converting a Claude Code plugin. Cursor's Marketplace offers official plugins with one-click install and OAuth, and team admins can distribute MCP servers through a team marketplace ([Cursor docs](https://cursor.com/docs/mcp)). *Use a plugin when* you want distribution and polish inside one app. Underneath, it's still MCP and skills, so if you need portability, ship the MCP server too. **The cheat sheet** | Need | Reach for | |---|---| | Live data / take actions | MCP server | | Repeatable workflow or house rules | Skill | | One-click install inside one app | Plugin / connector | | Works across Claude, ChatGPT, Cursor | MCP server (+ skill) | **Gotchas that bite** - **Context bloat:** every MCP tool you enable is another option the agent has to weigh. Cursor lets you toggle servers off without removing them, and enterprise admins can allowlist servers and specific tools. - **Skills don't sync:** custom skills on claude.ai, the Claude API and Claude Code are managed separately. Upload once per surface. - **Sandbox differences:** skills on the Claude API run with no network access and no runtime package installs. In Claude Code they get your machine's network. - **Trust:** both Anthropic and Cursor say the same thing: only install skills and servers from sources you trust, and audit the code. A skill can run scripts. An MCP server can act on your behalf. Treat both like installing software. **Chef's pick:** start with one remote MCP server for the system you touch most (GitHub, your DB, your tracker), then write one skill that says *how* your team uses it. Hands plus recipe beats a drawer full of gadgets. 🍽️ --- ## Secret Sauce 🥫 **Try it now — Claude Code idle compaction (v2.1.299+):** 1. Update Claude Code to v2.1.299 or later. 2. Idle 200K+ token terminal chats now compact after ~54 minutes, before the 1-hour prompt cache expires — so step away guilt-free. 3. Want the full transcript kept? Set `"idleCompaction": false`, or raise the threshold with `CLAUDE_CODE_IDLE_COMPACT_MIN_TOKENS`. --- ## Video Tasting 📺 **[Claude Agent Skills Explained](https://www.youtube.com/watch?v=fOxC44g8vig)** · 🟡 Intermediate · 3:14 · Anthropic (official) **After watching you can:** Tell skills, CLAUDE.md, MCP servers and subagents apart, and know when to use each. - Skills are folders of expertise Claude loads automatically when a task matches. - Only each skill's name and description load at startup; the full SKILL.md loads on demand. - Compares skills with CLAUDE.md (project facts), MCP (connects to data) and subagents (fixed roles). **Short on time? Watch just:** [0:30](https://www.youtube.com/watch?v=fOxC44g8vig&t=30s) How Agent Skills work · [1:42](https://www.youtube.com/watch?v=fOxC44g8vig&t=102s) Agent Skills vs MCP Servers · [2:33](https://www.youtube.com/watch?v=fOxC44g8vig&t=153s) Stacking skills into one bigger agent **On screen:** mostly a code editor: skill files, CLAUDE.md, an mcp.json with Playwright, and skills stacked into one bigger agent. [Summary](#video-tasting) · [Anthropic](https://www.youtube.com/channel/UCrDwWp7EBBv4NwvScIpBDOA). More videos: [/videos/](/videos/) --- ## New on the Shelf 🛒Fresh MCP servers, skills, plugins and templates, each checked against its primary source.