Choosing an AI agent shouldn't take a week of tabs.
AgentGid tracks 250+ of them with live public data:
- find yours in 3 questions
- compare any two side by side
- see if it's down right now
- what changed this month
- 40K MCP servers, ranked
Free: agentgid.com
we read the configs of 9,964 public GitHub repos set up for AI coding agents.
then counted every MCP server inside.
these 12 are what devs actually plug in. not hype. config files.
01 Context7 → up-to-date library docs for your agent · 15.1% github.com/upstash/context7
02 Playwright MCP → your agent drives a real browser · 13.6% github.com/microsoft/playwri…
03 GitHub MCP → issues, PRs and code without leaving chat · 12.1% github.com/github/github-mcp…
04 Supabase → your database and auth · 7.4% github.com/supabase/mcp
05 Filesystem → read and write only the folders you allow · 6.6% github.com/modelcontextproto…
06 Sequential Thinking → step-by-step problem solving · 6.1% github.com/modelcontextproto…
07 shadcn/ui → browse and add UI components · 4.8% github.com/shadcn-ui/ui
08 Chrome DevTools → console, network, performance · 4.6% github.com/ChromeDevTools/ch…
09 Memory → knowledge-graph memory across sessions · 4.6% github.com/modelcontextproto…
10 Microsoft Learn → official Microsoft docs · 3.3% github.com/MicrosoftDocs/mcp
11 Serena → semantic code search and edits · 2.7% github.com/oraios/serena
12 Task Master → task lists your agent works through · 2.7% github.com/eyaltoledano/clau…
#1 combo we found: Claude Code + Playwright MCP + Context7, together in 273 repos.
docs → browser → repo → data → memory → plan.
% = share of the 8,374 repos with an MCP config. full breakdown by agent and project type, updated weekly: agentgid.com/stacks/real-wor…
bookmark it before you set up your next agent ↓
1
2
63
Teardown: A Claude Code Skill That Builds 3D Machine Simulators
A viral X article shows teardown, a free Claude Code skill that builds a machine as a 3D simulator in one offline HTML page and records it to video.
agentgid.com/guides/teardown…
@claudeai
21
Free assets rising fastest this week, by downloads:
1. Revideo: programmatic video rendering (+211%)
2. Avvvatars: avatar generator for React (+32%)
3. PMNDRS Assets: fonts, HDRIs, models, textures (+27%)
Top 10 + new this week: agentgid.com/top/week/#asset…
@pmndrs
1
22
Fastest-rising APIs last week, by official SDK downloads:
1. Cartesia: text-to-speech and voice agents (2.1×)
2. ScrapeGraphAI API: LLM-powered web scraping (1.7×)
3. Messaging API: LINE messaging for agents (1.6×)
Top 10 + new this week: agentgid.com/top/week/#apis
@cartesia
36
Agent skill and plugin that turns a video URL or local file into timestamped frames and a transcript so Claude Code, Codex and other Agent Skills hosts can analyze it. agentgid.com/agents/claude-v…
27
Grab's designers and ops teams ship their own fixes with Cursor
Non-engineers get value from a coding agent fastest on concrete bugs with a visible check, as long as they learn Git…
Source: Cursor customer story.
agentgid.com/use-cases/grab-…
@cursor_ai
27
Your security patch is becoming an attacker’s roadmap. 😬
Google’s latest report describes how attackers are using AI agents to:
Spot a newly released security update.
Compare the old code with the patched version.
Identify the vulnerability the patch fixed.
Generate an exploit.
Attack systems that haven’t installed the update yet.
Google has observed these pipelines running without a human in the loop. Google Cloud Blog
The window to patch is shrinking. “We’ll update later” is becoming a dangerous bet.
1
1
37
Full research cloud.google.com/blog/topics…
22
Did none of these LLMs play Spec Ops: The Line? 😺🤪😱
Vals AI tested leading models on battlefield targeting tasks. Apparently, the safety settings have a gaming mode.
• Models flagged civilians, medics and surrendering soldiers as targets.
• GPT-6 Astra selected at least one target in 6% of responses. Claude Sonnet 5.5: 36%.
• Tell them “it’s just a game,” and those numbers jump to 92% and 82%.
• In follow-up scenarios, most models kept going even after being shown simulated casualties. Vals AI
Then comes the school test.
Using real satellite imagery of Minab, Iran, Claude Opus 5.5 marked the school as a target in 22 out of 32 trials. With game framing: 23 out of 32.
The higher-resolution image even showed a playground with children. Vals AI
These were target-identification tests, not models independently ordering strikes. But “just tell it you’re gaming” is a pretty disturbing weakness in the safeguards.
No conclusions. Someone send Claude a copy of Spec Ops. 😬😬😬
Full study: vals.ai/blogs/autonomous-war…
2
55
Libraries devs suddenly reached for last week (vs usual):
1. Desktop Commander: terminal and file editor (2.5×)
2. Vercel Postgres: edge Postgres client (2.4×)
3. vinext: Next.js apps on Vite (2.4×)
Top 10 + new this week: agentgid.com/top/week/#tools
@vercel
1
39
LoanPro built a loan servicing assistant on AWS Strands Agents:
→ summaries in under 15 s vs 3–5 min by hand
→ call times down up to 15%
→ ~8% fewer internal tickets
→ Strands Agents: 8,913,553 weekly downloads
agentgid.com/stories/loanpro…
1
36
OMG!!! ‼️these 20 GitHub repos turn Fable 5.2 into a full design studio.
taste. systems. motion. canvas. assets. QA.
all from the AgentGid catalog. stars checked today.
01 UI UX Pro Max → UI styles, palettes, font pairings and a UX checklist before it builds a page. ★150k github.com/nextlevelbuilder/…
02 Taste Skill → picks one of 20 design directions, finds credited images, checks the page in a real browser. ★117k github.com/Leonxlnx/taste-sk…
03 Impeccable → a design director's rulebook and 24 commands. Catches AI slop after every edit. ★111k github.com/pbakaus/impeccabl…
04 OpenDesign → local design app that runs on your coding agent. Pages, decks, videos in your brand style. ★108k github.com/nexu-io/open-desi…
05 /frontend-design → Anthropic's own skill for pages with a distinct look. ★196k (whole repo) github.com/anthropics/skills
06 Unslop → turns vibecoded, AI-looking UI into a designed interface. ★17k github.com/heydamir/unslop
07 Emil Kowalski's skills → motion and polish. Knows when something should not move. ★57k github.com/emilkowalski/skil…
08 shadcn/ui → copy the component code into your project and make it yours. ★128k github.com/shadcn-ui/ui
09 Motion → animation for React and JavaScript. ★35k github.com/motiondivision/mo…
10 GSAP skills → official skills that teach your agent to use GSAP right. ★18k github.com/greensock/gsap-sk…
11 Three.js → 3D in the browser. ★118k github.com/mrdoob/three.js
12 Excalidraw → whiteboard with a hand-drawn look. Sketch flows before you build. ★135k github.com/excalidraw/excali…
13 tldraw → build your own infinite-canvas app in React. ★51k github.com/tldraw/tldraw
14 Penpot → open-source design and prototyping platform. ★62k github.com/penpot/penpot
15 Archify → idea or codebase into an interactive diagram, one HTML file. ★97k github.com/tt-a1i/archify
16 HyperFrames → write HTML, render MP4. Preview first, render after you approve. ★85k github.com/heygen-com/hyperf…
17 Lucide → 2,000+ open-source icons in one consistent style. ★25k github.com/lucide-icons/luci…
18 Google Fonts → the font files behind fonts.google.com. ★20k github.com/google/fonts
19 rembg → remove image backgrounds locally. ★25k github.com/danielgatis/rembg
20 agent-browser → browser automation CLI built for agents. Let it look at what it just built. ★52k github.com/vercel-labs/agent…
taste → system → motion → canvas → ship.
💡💎❗️save this. license, install line and what each one actually does: agentgid.com
Design and media assets on the rise this week:
1. Revideo: programmatic video rendering (+63%)
2. Pixelarticons: pixel art SVG icons (+44%)
3. Tegaki: handwriting animation library (+28%)
Top 10 + new this week: agentgid.com/top/week/#asset…
@halfmage @gkurttech
1
2
92
APIs builders wired in more last week (SDK downloads):
1. Triton One: Solana RPC and gRPC streaming (2.6×)
2. Supabase (pgvector): Postgres with vector search (2.3×)
3. Parallel: web search and research (2.1×)
Top 10 + new this week: agentgid.com/top/week/#apis
@triton_one
54
Doctolib got 47% more pull requests per engineer with Claude Code
When coding agents make writing code cheap, the next gains come from automating review, risk checks and post-merge verification around each change…
Source: Anthropic customer story.
agentgid.com/use-cases/docto…
1
48
Last week's breakout libraries, vs their usual week:
1. Supabase Server: Supabase auth utilities (7.5×)
2. context-mode: context window optimizer (3.0×)
3. mcp-remote: remote MCP server connector (2.9×)
Top 10 + new this week: agentgid.com/top/week/#tools
1
35
GPT-6.1 Sol scores within 2.7 points of the top model on Gid Bench for about 5x less per task. Gid Bench: 13 public benchmarks in 4 groups, updated daily. Data: Epoch AI (CC BY 4.0), prices: LiteLLM. agentgid.com/benchmarks/char…
2
35












