As AI coding tools proliferate across your engineering org, a critical governance gap is emerging: developers are spinning up MCP servers faster than platform teams can audit them.
As AI coding tools proliferate across your engineering org, a critical governance gap is emerging: developers are spinning up MCP servers faster than platform teams can audit them. Week 37 reveals which integrations your team is already using—and what security decisions you need to make about them.
No new servers entered the CuratedMCP catalog this week, but that's not a slowdown—it's a sign of maturity. The existing 72 risk-classified servers are seeing accelerating adoption. This is the moment to ensure your allowlist policies are enforced consistently across Claude Code, Cursor, Windsurf, and GitHub Copilot. If you haven't conducted a live audit of which servers are running on developer machines right now, Week 37 is your reminder: shadow MCP usage is real, and it's growing.
The five most-viewed servers this week tell you exactly what your developers want to connect to their AI agents:
GitHub Copilot MCP (98K views) and GitHub MCP (76K views) dominate. Both grant AI agents deep access to your source code, pull requests, and workflows. Governance consideration: Does your team require mutual TLS or IP allowlisting for these integrations? Are audit logs being retained?
OpenAI MCP (87K views) lets developers route requests through OpenAI's APIs from within Claude or other clients. Supply-chain risk: You're now dependent on two LLM vendors' uptime and data handling policies. Token spend visibility becomes essential when requests can flow through multiple providers.
Figma MCP (82K views) exposes design tokens and component libraries to AI agents. AppSec angle: Who owns the Figma workspace credentials being stored locally? Is there a credential rotation policy?
Anthropic Claude MCP (76K views) allows nesting Claude as a sub-agent. This is architectural—it adds latency, multiplies token consumption, and creates audit visibility challenges across multiple inference calls.
Here's what keeps platform leaders awake: you cannot govern what you cannot see. This week's view counts reflect organic developer demand, but they likely undercount actual deployments. Developers are running MCP servers across four separate IDE integrations (Claude Code, Cursor, Windsurf, GitHub Copilot), each with its own local configuration. Your RBAC and allowlist are only as strong as your ability to enforce them per-machine.
Start here: audit which MCP servers are actually running in your org right now. Use CuratedMCP's per-machine enforcement to lock down policy—don't rely on developer best practices. Then, layer in visibility: when developers connect OpenAI MCP or Figma MCP, you should know about it the same way you monitor SaaS app usage. If you're seeing multiple Claude integrations and nested Claude calls, you're also seeing token spend sprawl. TokenShield gives you the visibility to measure that spend and understand which optimization patterns actually work in your environment—before you commit to cost-saving approaches that might degrade your team's velocity.
The policy library at 72 risk-classified servers means you have a foundation. Now enforce it.
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