Dispatch #1: what actually changed in AI coding tools this cycle

Windsurf isn’t Windsurf anymore

If you’re comparing AI coding tools off an article from last year, this is the correction to make first: Windsurf, as a standalone brand, has been folded into a product called Devin Desktop. Navigating to windsurf.com now redirects straight to Cognition’s devin.ai/desktop, and that page says so directly — “Devin Desktop is the new name for Windsurf.” Existing Windsurf users’ plans, extensions, and settings carried over automatically, and a separately maintained “Windsurf for JetBrains” plugin still exists under the old name (devin.ai/desktop). Full breakdown of what that means for how the product is positioned now — a fleet-of-agents dashboard rather than an editor-with-AI — in our Cursor vs Claude Code vs Windsurf comparison.

GitHub Copilot’s top tier includes… Claude Code

Genuinely worth flagging: GitHub Copilot’s Pro+ tier ($39/month) and Max tier ($100/month) both list “third-party agents” as an included capability, and GitHub names them explicitly — “Claude Code, Codex.” (github.com/features/copilot/plans). If your team is already paying for Copilot Pro+ or Max, you may already have Claude Code access bundled in before you buy it separately. We go through the full pricing table and where the two tools’ architectures actually diverge in Claude Code vs GitHub Copilot.

The MCP “official servers” list has moved — check before you copy a tutorial

A lot of still-indexed “here are the official MCP servers” posts point at a repo layout that’s no longer accurate. As of writing, the servers actively maintained by the MCP Steering Group are: Everything, Fetch, Filesystem, Git, Memory, Sequential Thinking, and Time. GitHub, PostgreSQL, and Slack — all commonly cited in older posts — have moved to a separate servers-archived repo (modelcontextprotocol/servers). The current community catalog lives at the MCP Registry instead. Our MCP servers review has the fuller list of what’s actually worth wiring up, plus the security question worth asking before you approve any server’s install command.

Worth re-reading if you haven’t

• Getting real work out of AI agents: a practical setup — the verification-loop and hooks pattern that separates an agent that looks productive from one that actually is.
• Claude Code, honestly — current pricing and the honest caveats, re-checked this cycle.

That’s Dispatch #1. Back issues and the signup form live at /dispatch/.