Airsc is free for everyone. If it helps you, support it with $2–$20.

Support
VISign in

New AI agent resources this week (week 43, 2026)

Updated 19/10/2026

This digest is for vibe coders who want to try new tools without reading every README. Everything here was added to Airsc in the last 7 days. Most picks have a 'safe' label and a quality score of 90 or higher.

This week leans toward design output, diagrams and agent memory. You can get better-looking pages, PDFs and architecture charts from your agent. You can also keep context across Claude, Cursor and ChatGPT, and build better debugging habits.

  1. PluginSafe

    taste-skill

    Leonxlnx

    Gives coding agents 'taste' by providing art direction, reference images, and automated browser-based visual checks.

    99kMIT

    #1 · Why we picked it

    Safe, quality 95, beginner level, and it checks UI changes in a real browser so the agent does not ship a generic layout.
  2. PluginSafe

    next-ai-draw-io

    DayuanJiang

    An MCP server that lets your agent draw and edit professional draw.io architecture diagrams and flowcharts using simple chat commands.

    36kApache-2.0

    #2 · Why we picked it

    Safe, quality 95, beginner level, and it draws draw.io architecture diagrams and flowcharts from plain chat commands.
  3. PluginSafe

    Kami

    tw93

    Layout rules and templates that allow AI agents to design and export professional PDF, PNG, and editable PowerPoint documents.

    12kMIT

    #3 · Why we picked it

    Safe, quality 95, beginner level, and it exports PDF, PNG and editable PowerPoint, such as a resume from Markdown.
  4. PluginSafe

    Waza

    tw93

    A suite of specialized commands like /think and /hunt that teach AI agents disciplined engineering habits for planning, debugging, and UI design.

    7.3kMIT

    #4 · Why we picked it

    Safe, quality 95, beginner level, and commands like /think and /hunt push the agent to plan and debug in a disciplined way.
  5. MCP serverSafe

    second-brain-cloudflare

    rahilp

    A self-hosted memory layer on Cloudflare that shares context across Claude, ChatGPT, and Cursor.

    806MIT

    #5 · Why we picked it

    Safe, quality 95, and a self-hosted memory layer that shares context across Claude, ChatGPT and Cursor; it needs a Cloudflare account and setup.
  6. MCP serverSafe

    erd-editor

    dineug

    Visual database schema designer with an MCP server that enables AI agents to create and edit entity-relationship diagrams live.

    1.7kMIT

    #6 · Why we picked it

    Safe, quality 90, beginner level, and lets an agent design and edit database diagrams live, then generate SQL DDL.
  7. MCP serverSafe

    world-intel-mcp

    marc-shade

    Provides 139 MCP tools for real-time global intelligence on markets, geopolitics, and environmental data without requiring paid API keys.

    909MIT

    #7 · Why we picked it

    Safe, quality 92, beginner level, with 139 tools for markets and world events that do not need paid API keys.
  8. MCP serverCaution

    Trail-Code

    trailhq

    An MCP server that optimizes agent context, reducing token costs and improving accuracy through codebase mapping.

    10kMIT

    #8 · Why we picked it

    Quality 92 and popular, it can cut token use with a compressed codebase map, but it carries a 'caution' label, so review its hooks before use.

How to choose

  • Start with beginner-level items marked 'safe'. Most of this week's top picks fit that.
  • Match the tool to your current task: taste-skill or Kami for visuals, next-ai-draw-io or erd-editor for diagrams, Waza for debugging habits.
  • Check setup needs first. Memory servers, SMS gateways and GitLab tools need accounts, keys or your own hosting.
  • Treat 'caution' items such as Trail-Code and video-talkcraft with care. Read what they run and test in a throwaway project first.

FAQ

Which resource is best for a first-time user this week?

taste-skill, Kami and next-ai-draw-io are all safe, beginner level and quality 95. Pick the one that matches what you build most often: pages, documents or diagrams.

Do any of these need API keys or extra setup?

Yes. second-brain-cloudflare needs a Cloudflare setup, gitlab-mcp needs access to GitLab, and textbee needs an Android phone as an SMS gateway. world-intel-mcp is listed as working without paid API keys.

What does the 'caution' label mean here?

It means the resource needs a closer look before you install it. Trail-Code and video-talkcraft have this label, so read their code and permissions and try them in a test project first.

Are there picks for specific stacks?

Yes. kotlin-footguns targets Kotlin and Compose Multiplatform, and gitlab-mcp suits teams on GitLab. Both are safe and intermediate level.