Week 45, updated 2 Nov 2026

AI use cases Agent skills & MCP

Agent skills & MCP

Add-ons that give an AI agent new abilities: Agent Skills (instruction packs an agent loads when needed) and MCP servers (connectors to tools like GitHub, a browser or up-to-date docs).

Tracked tools
9
Highest score
62
Caution labels
1
Use cases
4
Agent skills & MCP, Week 45score 0–100, safety
  1. 1Superpowers#13 overall · Agent skill62No dependencies scanned
  2. 2Context7#19 overall · MCP server57Caution
  3. 3Agent Skills#25 overall · Agent skill53No issues found
  4. 4MCP reference servers#29 overall · MCP server48No issues found
  5. 5awesome-mcp-servers#30 overall · Directory47No dependencies scanned
  6. 6Playwright MCP#32 overall · MCP server45No issues found
  7. 7wshobson/agents#38 overall · Agent skill42No issues found
  8. 8GitHub MCP Server#40 overall · MCP server41No issues found
  9. 9skill-recorder#52 overall · AI tool22No issues found

What you can do

Add-ons that give an AI agent new abilities: Agent Skills (instruction packs an agent loads when needed) and MCP servers (connectors to tools like GitHub, a browser or up-to-date docs). For people already using Claude Code, Codex, Cursor or similar who want their agent to do more or follow better workflows.

Pick by situation

Based only on measured data for the 9 tracked tools here.

  • SuperpowersFits if you want the highest overall score (score 62, #13 of 52 overall)
  • awesome-mcp-serversFits if active maintenance matters most (last push 2 Nov 2026, maintenance 37)
  • SuperpowersFits if you want the most widely adopted option (310,769 stars, 27,654 forks)
  • SuperpowersFits if you want the most thorough README docs (docs score 69, 14,835-character README)
  • MCP reference serversFits if you need a permissive license and a clean scan (Apache-2.0 / MIT, no issues found by automated scan)

Short answer

Of the 9 tools we track in Agent skills & MCP, the highest scored this week are Superpowers (62), Context7 (57), Agent Skills (53), MCP reference servers (48) and awesome-mcp-servers (47). Safety labels in this category: 6 no issues found, 2 no dependencies scanned, 1 caution. “No dependencies scanned” (Superpowers, awesome-mcp-servers) means the code checks found nothing but there was no runtime lockfile to check. An MCP server or agent skill runs with your agent's permissions, so we scan each one weekly: dependencies, install hooks, agent instruction files and, for MCP servers, whether the code spawns processes, touches files or uses the network.

Read the full answer

The MCP servers we track

  • Context7: score 57, Caution (2 critical dependency advisories). Serves up-to-date library documentation to LLMs and AI code editors via MCP.
  • MCP reference servers: score 48, No issues found; our MCP code check found: code shell/process exec, filesystem read, filesystem write and network. Reference collection of Model Context Protocol (MCP) servers from the MCP project.
  • Playwright MCP: score 45, No issues found. Microsoft's MCP server that lets agents control a browser through Playwright.
  • GitHub MCP Server: score 41, No issues found. GitHub's official MCP server for repos, issues and pull requests.

The same ranking with setup notes per client: MCP servers for Claude Code · MCP servers for Codex · MCP servers for OpenCode

MCP server security: what to check before you connect one

  • A local MCP server is code you run. Without sandboxing and consent, a malicious startup command or server payload can execute “any command with MCP client privileges”.
  • Your client should show the exact command it will run before adding a one-click server, and require explicit approval.
  • Prefer the stdio transport for local servers, or require an authorization token if a local server uses HTTP.
  • Sandbox servers with minimal default access to files and network.

Source: MCP security best practices. For Claude Code users: servers in a project's .mcp.json get their own approval prompt, and Anthropic “does not security-audit or manage any MCP server” (Claude Code security docs).

What to watch out for

Scores are relative public signals; none of these tools were tested hands-on.

  • License to check before commercial use: Agent Skills (Apache-2.0 + proprietary).
All scan reasons
  • Superpowers: code and instruction-file checks found nothing, but there is no runtime lockfile or manifest, so dependencies were not checked
  • Context7: 2 critical-severity known vulnerabilities in runtime dependencies (OSV)
  • awesome-mcp-servers: code and instruction-file checks found nothing, but there is no runtime lockfile or manifest, so dependencies were not checked

Full table

Open the sortable table
Agent skills & MCP: highest scored first
Tool (overall rank)Use caseScoreSafetyLast pushStars
13 Superpowersobra/superpowersA skills framework and software-development methodology for coding agents. Agent skills & MCP62 No dependencies scanned★ 310,769
19 Context7upstash/context7Serves up-to-date library documentation to LLMs and AI code editors via MCP. Agent skills & MCP57 Caution★ 64,060
25 Agent Skillsanthropics/skillsAnthropic's public repository of Agent Skills, reusable instruction packs that agents load on demand. Agent skills & MCP53 No issues found★ 186,457
29 MCP reference serversmodelcontextprotocol/serversReference collection of Model Context Protocol (MCP) servers from the MCP project. Agent skills & MCP48 No issues found★ 91,744
30 awesome-mcp-serverspunkpeye/awesome-mcp-serversLarge community-curated list of MCP servers. Agent skills & MCP47 No dependencies scanned★ 97,179
32 Playwright MCPmicrosoft/playwright-mcpMicrosoft's MCP server that lets agents control a browser through Playwright. Agent skills & MCP45 No issues found★ 38,961
38 wshobson/agentswshobson/agentsPlugin marketplace of agents and skills for Claude Code, Codex, Cursor, OpenCode and more. Agent skills & MCP42 No issues found★ 42,088
40 GitHub MCP Servergithub/github-mcp-serverGitHub's official MCP server for repos, issues and pull requests. Agent skills & MCP41 No issues found★ 34,149
52 skill-recordermicrosoft/skill-recorderDesktop app that records a work session and turns it into a reusable agent Skill or automation. Agent skills & MCP22 No issues found★ 5,640

Score = relative rank (percentile) among all 52 tracked tools on maintenance, adoption and docs. Community signal is not measured yet. How we score.

Get the board every Monday

One email a week: the new top 10, the biggest score changes and every safety label that changed.

Free. Double opt-in: you confirm by email first, unsubscribe in one click.