CoolDeep AI Digest

The author realized that for two years they had been treating Claude like a simple search engine—opening a new chat, pasting context, asking a question, copying the answer, and closing the tab—while missing most of the platform’s power. A “100 Claude Tips” image revealed features they had never used: Projects, Skills, MCP Connectors, Memory, Claude Code, Prompt Caching, and the Batch API. Recognizing that they were only scratching the surface, they spent days testing these capabilities and now share a step‑by‑step guide to move from disposable chats to a compounding AI workflow. **Key details and examples** - **Projects**: Inside Claude, a Project stores brand voice, tone guidelines, reference documents, and other context once. Every conversation in that Project automatically inherits that information. The author created a CoolDeep AI Project with newsletter voice guidelines and past issues, so each draft now matches their style from the first line. - **Skills**: A Skill is a saved workflow that encodes repeated instructions (e.g., “write in my voice,” “keep under 300 words,” “give three options”). By going to Claude Cowork → Customize → Skills → Add and describing the workflow, Claude builds and tests the skill file, eliminating the need to repeat those prompts. - **MCP Connectors**: These let Claude access external tools such as Google Drive, Gmail, GitHub, Slack, and the web via the “Claude in Chrome” connector. Instead of relying only on pasted text, Claude can pull real data, read emails, browse repositories, or search the internet in real time. - **Memory**: Enabling “Generate memory from chat history” in Settings lets Claude build a longitudinal profile of the user’s preferences, work style, and projects. Over weeks, the assistant begins to anticipate needs without explicit reminders. - **Claude Code**: For developers or anyone writing scripts, Claude Code is a terminal‑based tool that edits multiple files, runs CI pipelines, handles debugging, and works across whole projects—far more capable than the chat interface for serious coding. - **Prompt Caching & Batch API**: When using the Claude API, Prompt Caching stores repeated system prompts or reference documents, cutting input costs by up to 90 %. The Batch API processes non‑urgent jobs at a 50 % discount with a 12‑ to 24‑hour turnaround, ideal for bulk content generation or classification tasks. **Actionable takeaways** 1. Create a Claude Project for each recurring domain (newsletter, research, coding) and upload relevant style guides and reference material. 2. Identify any prompt or workflow you use more than twice and turn it into a Skill via Claude Cowork. 3. Connect Claude to the tools you use most (Drive, Gmail, GitHub, Slack) through MCP connectors; start with the web‑browsing connector for live research. 4. Turn on memory generation in Settings so Claude learns your habits over time. 5. If you code, install Claude Code and use it with Claude Sonnet 4.6 for file‑level edits and pipeline work. 6. For API‑based work, enable Prompt Caching for repetitive inputs and submit large, non‑time‑sensitive jobs to the Batch API to reduce costs. By adopting these practices, the user shifts from asking isolated questions to building a persistent, context‑aware AI assistant that compounds value, saves time, and lowers expenses.