Originally published at curatedmcp.com/blog/week-2026-31

MCP Ecosystem Week 31: When Official Integrations Become Your Biggest Allowlist Decision

The MCP ecosystem continues to consolidate around official integrations from major vendors — and that consolidation is exactly where governance matters most. This week, zero new servers were reviewed, but the usage data tells a clearer story: your developers are converging on a small set of high-trust integrations. That's good for security surface area. It's bad if you haven't formally decided whether to allow them.

This Week in MCP

No new servers entered the catalog this week, but that's not stagnation — it's signal that platform teams are focusing on vetting and deploying the existing 74 risk-classified servers rather than chasing novelty. If you're running CuratedMCP, this is the moment to audit which of those 74 are actually in use across your developer fleet, versus which are sitting in your policy library untouched.

The real governance work happens now: moving from "we have a catalog" to "we have a decision for every server our developers want."

On the Radar

The five most-viewed servers this week are all official integrations from major platforms, and they warrant explicit allowlist decisions:

GitHub Copilot MCP (98K views) — Direct integration with GitHub Copilot's code intelligence. Before allowlisting: confirm your SSO grants the right GitHub org scope, and audit whether Copilot telemetry is routed through your compliance boundary.

OpenAI MCP (87K views) — Access to GPT-4o, DALL-E, Whisper, and Embeddings. This is a supply-chain node: it requires OpenAI API keys. Enforce key rotation policy and segregate API spend across teams if you're tracking it.

Figma MCP (82K views) — Design file and token access in your AI workflow. Data classification concern: confirm whether design files are appropriately scoped in Figma's RBAC, and whether your agents should have access to all components or a subset.

GitHub MCP (76K views) — Repo, issue, PR, and workflow management. This is a high-impact surface: agents writing to repos need audit trails. Enforce branch protection and require pull request review even for AI-generated code.

Anthropic Claude MCP (76K views) — Nested Claude reasoning. Governance question: are you comfortable with intra-Claude API calls, and do you want to meter them separately from first-party Claude usage?

Governance Take

Here's the real risk this week: allowlist fragmentation across your IDE and agent fleet.

You've likely deployed Claude Code, Cursor, Windsurf, and GitHub Copilot across your team. Each has its own MCP integration pathway. Each developer can independently add servers. And unless you're actively enforcing policy at the machine level — not just documenting it — you have no view into which servers are actually running in which client.

The five most-viewed servers aren't controversial individually. But "GitHub Copilot MCP is allowed in Cursor" and "GitHub Copilot MCP is blocked in Claude Code" is a policy debt that compounds weekly. Platform teams we work with find, on audit, that their allowlist has drifted across clients within months.

Second: token spend and audit-log visibility aren't the same thing. With TokenShield, you get a live ledger of Claude spend broken down by machine and user. That visibility is the foundation for governance — you can see where your spend is concentrating, and whether it maps to your allowlist decisions. If 60% of Claude tokens are flowing through a server you haven't reviewed, that's a policy gap. TokenShield surfaces that gap in days, not months.

This week: run an audit across your fleet. Log into CuratedMCP and compare your active allowlist against actual usage. The gap is where governance happens.

---

Govern MCP usage across your team with CuratedMCP — or scan your own stack free at https://www.curatedmcp.com/auditor.