MCP is winning quietly — 4,000 servers and the integration combinatorics problem is solved
The Model Context Protocol crossed 4,000 published servers in May. The network effect is now the lock-in. The only open question is whether any vendor still tries to fragment it.
The 4,000 milestone
The MCP server registry crossed 4,000 published servers — roughly a 6× growth since January. Most are open-source and community-maintained. The combinatorics problem (every tool × every agent = N×M custom integrations) is solved by the network effect, not by central planning.
The economic asymmetry
An agent tool shipping with MCP support inherits ~4,000 integrations for free. An agent tool shipping with a proprietary integration framework starts from zero. That cost asymmetry pushes every serious vendor into MCP support, which compounds the registry growth, which compounds the asymmetry.
This is the textbook winner-take-most network effect. Once the protocol crosses a critical mass of supply (integrations) and demand (agents that use them), the equilibrium becomes nearly impossible to displace without an order-of-magnitude better protocol — and nobody is currently shipping one.
The Copilot question
The only remaining strategic question is whether GitHub Copilot moves from partial MCP support to full. Partial support fragments the standard — Copilot users get a different integration tier than MCP-native users. Full support consolidates MCP as a true cross-vendor standard.
Microsoft's MCP decision is the load-bearing strategic moment for whether the standard hardens or splinters.
The case for fragmenting is the GitHub-and-VS-Code ecosystem is large enough that Microsoft can credibly run its own integration framework and force enterprise procurement teams to support both. The case for consolidating is the broader Cursor + Windsurf + Claude Code + Antigravity 2.0 tier already supports MCP, and Microsoft losing the "our framework or theirs" argument across that field would be expensive.
The procurement read
For enterprise buyers, MCP support should now be a baseline qualifier in any agent-tooling RFP. Vendors that ship partial support are signaling — either lack of ecosystem maturity or a strategic intention to push customers into proprietary integrations later. Neither signal favors the buyer.
If you're standing up agent tooling today, the safe bet is the full-MCP tier. The deferred decision is whether Copilot's partial support hardens into a credible alternative or capitulates. Either way, the buyer who optimized for full-MCP coverage in 2026 is the buyer with the cleanest integration surface in 2027.
Cosmic JS — Claude Code vs Copilot vs Cursor → · SD Times — May 8 2026 AI → · Lushbinary — AI coding agents 2026 →