This week brought no new catalog entries, but the usage signal is loud: developers are converging on official integrations. The top five most-viewed servers—GitHub Copilot, OpenAI, Figma, GitHub, and Anthropic Claude MCPs—collectively represent over 419,000 views.
This week brought no new catalog entries, but the usage signal is loud: developers are converging on official integrations. The top five most-viewed servers—GitHub Copilot, OpenAI, Figma, GitHub, and Anthropic Claude MCPs—collectively represent over 419,000 views. That's not just demand; it's a governance inflection point. If your team is running Claude in Cursor, Windsurf, or VS Code, and developers are pulling in these servers unsupervised, you're seeing live supply-chain expansion happen in real time.
The quiet week in new catalog submissions also signals maturation. The ecosystem isn't exploding in breadth anymore; it's consolidating around high-signal integrations. For platform teams, that means the blocking/allowlisting surface is getting smaller but the stakes are higher.
No new servers were reviewed and risk-classified this week. The catalog remains at 73 total risk-classified servers, 72 free-to-use. The lull in new submissions reflects a natural rhythm: teams are focusing on hardening existing integrations rather than racing to add every tool under the sun.
The stability is good news for governance. It gives your platform team room to breathe, audit existing allowlists, and tighten policies around the servers developers actually use.
The five most-viewed servers warrant close attention from any platform or AppSec team rolling out AI coding tools:
GitHub Copilot MCP (98,010 views) and GitHub MCP (76,016 views) grant AI agents direct access to repositories, pull requests, and workflows. Governance question: who authenticates, and does your token rotation policy cover service accounts? If one dev's Cursor session leaks a GitHub PAT, how far does the blast radius extend?
OpenAI MCP (87,005 views) opens GPT-4o, DALL-E, and Embeddings access. This is spend multiplication without visibility—unless you're monitoring token usage per machine and per user. Blind spots here can balloon costs across your organization.
Figma MCP (82,008 views) exposes design files and tokens in your coding workflow. Data classification question: is every design file supposed to be accessible to every developer's AI agent? Does your Figma workspace permission model map cleanly to your IDE-level access controls?
Anthropic Claude MCP (76,004 views) nests Claude-as-a-sub-agent, which is powerful for specialized tasks but multiplies token spend and audit complexity. You now have multiple Claude calls per prompt, sometimes in a chain, sometimes recursive.
Here's the sharp reality: IDE-native MCP adoption is outpacing your audit infrastructure. Developers install Claude in Cursor, add three MCPs from the community, and your allowlist—built around Slack integrations and cloud SDKs—suddenly doesn't apply. You have no signal.
Start here: run an audit across your machines this week. Use the free auditor tool to scan which servers are actually running. Compare that list to your official policy. The gap is your governance debt.
Second: if you're using OpenAI MCP or other spend-multiplying integrations, you need token visibility. You can't govern what you don't measure. TokenShield gives you a live ledger of Claude spend per machine, per user, and per server—visibility that pairs with policy enforcement.
The five most-viewed servers are all official integrations, which is good—lower supply-chain risk than random community tools. But official ≠ automatically approved for your org. GitHub access, OpenAI costs, and design-file exposure all require explicit decision-making, not passive defaults.
Govern MCP usage across your team with CuratedMCP — or scan your own stack free at https://www.curatedmcp.com/auditor.
Explore the full MCP catalog
Discover, compare, and install verified MCP servers