The Weekly Index #1 — 11,394 Tools, 1,190 MCP Servers, and the Database Tool You Should Install Today

Welcome to issue #1 of The Weekly Index — the Thursday digest you signed up for. Every week: what's new in AI agent tools, what went stale, what you should actually install. No hype, no sponsorships.

📊 This Week's Numbers

• 11,394 tools indexed across 5 ecosystems
• 1,190 MCP servers (awesome-mcp-servers curated list)
• 1,818 GPT Actions
• 2,471 OpenClaw Skills
• 1,341 VS Code AI extensions
• Security note: 52.8% of scored tools in our index have a security score of 2/5 or below. We track this so you don't have to.

🔦 Spotlight: MCP Toolbox for Databases (Google)

Score: 94/100 · 13,133 GitHub stars · Security: 5/5

This week's spotlight goes to Google's MCP Toolbox for Databases — the most comprehensive database connectivity layer in the MCP ecosystem right now. It supports PostgreSQL, MySQL, SQLite, AlloyDB, Cloud SQL, Spanner, and Bigtable from a single server. Maintained by Google's engineering team with 26 contributors and active weekly commits.

What sets it apart: the toolbox abstracts connection pooling, authentication, and schema discovery. You point it at your database and it figures out the rest. Most MCP database servers require you to write raw SQL. This one understands your schema.

One thing to watch: Google Cloud dependency for some backend types. If you're fully self-hosted, DBHub (91/100) is the stronger pick.

Full profile →

📈 Trend: MCP Is Winning the Protocol War

A quiet but significant signal this week: we now have 1,190 MCP servers in the curated awesome-mcp-servers list alone — a list that had ~500 entries six months ago. Meanwhile, our GPT Actions collection (1,818 tools) has seen zero net additions in the past 30 days. OpenAI's plugin ecosystem appears to have plateaued while the MCP ecosystem continues to compound.

The reason isn't technical superiority. It's distribution. Anthropic opened the MCP spec, IDE makers (Cursor, Windsurf, Cline) adopted it, and a flywheel started. Building an MCP server is now a standard way to expose your API to AI agents regardless of which model the user runs.

🆕 Top 5 New Tools Worth Your Attention

1. MCP Toolbox for Databases — Google's multi-database MCP server. Score: 94/100.
2. DBHub — Zero-dependency, token-efficient database MCP for Postgres, MySQL, SQLite. Score: 91/100.
3. Chrome DevTools MCP — Browser debugging for coding agents. 26k stars. Score: 100/100.
4. Firecrawl MCP Server — Web scraping and search for agents. 5.5k stars. Score: 91/100.
5. Claude Task Master — Structured task management for Claude agents. 25.5k stars. Score: 94/100.

⚙️ Use Case Corner: Natural-Language SQL in 10 Minutes

Here is the setup that takes the least time to produce the most value:

1. Install supabase-mcp if you're on Supabase, or DBHub for any other database
2. Add your connection string to Claude Desktop config
3. Ask Claude: "Show me the 10 users who signed up last week and how many sessions each had"

Claude will write and execute the SQL, return formatted results, and offer follow-up analysis. No SQL knowledge required from your end. The security model: both tools default to read-only access. Writes require an explicit override.

👥 Community Picks

• Supabase MCP Server (Official) — Most-discussed in the Supabase Discord this week. The official version finally landed and the community prefers it over the community fork.
• cognee — Knowledge graph MCP server. 12k stars and growing. Useful for agents that need persistent memory across sessions.

---

See you next Thursday. Browse the full index at skillsindex.dev/browse · Past issues