Source: GitHub Changelog — 2026-07-14
Summary
GitHub's June update for Copilot in Visual Studio (distinct from the separate VS Code changelog) centers on visibility and trust: a refreshed real-time usage/billing window, a new MCP server trust-validation layer, and general availability for the C++ modernization agent's MSVC upgrade scenarios. The MCP trust feature is the most notable piece — Visual Studio now fingerprints an MCP server's config and assets at startup and blocks silent changes until a human approves them.
Key Takeaways
- MCP trust validation: Visual Studio compares an MCP server's configuration and asset fingerprint against a known-good baseline every time it starts. If anything changed, a trust dialog forces explicit review before the server is allowed to run — closing a gap where a compromised or swapped MCP server could silently gain tool access.
- Usage tracking: a refreshed Copilot Usage window shows real-time consumption against the usage-based billing model, with proactive alerts as you approach your limit, when you hit it, and when overage billing kicks in.
- The C++ modernization agent's MSVC upgrade scenarios graduate from preview to GA, offered in two modes: Automated (end-to-end unattended upgrade) or Guided (review the assessment, plan, and each execution step before it runs).
- Next Edit Suggestions now predicts follow-up edits anywhere in the active file, not just near the cursor — a small but compounding change to how much of a file Copilot proactively rewrites per session.
- Available across all Copilot plans, from Free through Enterprise.
Reel Script
Hook: Every AI coding agent with MCP tool access is a potential supply-chain attack vector — and most editors have no idea if the MCP server they trusted yesterday is the same one running today. Visual Studio just closed that gap.
Core Concept: MCP — the Model Context Protocol — is how coding agents like Copilot reach out to external tools: databases, ticket systems, internal APIs. The problem is that once you've approved an MCP server once, most tools just keep trusting it forever, even if its config or the assets it serves quietly change later. That's the same class of risk as a browser extension that gets sold and starts injecting ads — the permission you granted doesn't shrink even if the thing behind it does. Visual Studio's fix is a fingerprint check: every time the MCP server starts, it hashes the config and assets and compares that against the last known-good version.
Hands-On: Think of the flow as three steps that now happen automatically on every launch. Step one: Visual Studio computes a fingerprint of the MCP server's current config and files. Step two: it compares that fingerprint against the baseline it trusted last time. Step three — and this is the actual new behavior — if the two don't match, it doesn't just log a warning, it blocks the server from running and pops a trust dialog that makes you look at what changed before you approve it again. That's the same idea as a checksum verification on a software download, just applied continuously to a tool an agent calls dozens of times a session instead of once at install time. Pair that with the new C++ modernization agent's Guided mode — which shows you the assessment and plan before each execution step — and the pattern across this whole release is the same: don't just automate the action, make the diff visible before it runs.
Takeaway: If your team has wired any coding agent into MCP servers you don't fully control the supply chain for, this is the feature to go copy, not just admire — a startup fingerprint check is cheap insurance against a very expensive mistake. Check what MCP trust model your own editor or agent harness actually enforces before you assume "we approved it once" still means anything.