Most business owners think of an AI coding assistant as something that lives inside a developer's editor. It reads code, suggests code, maybe runs a test. That picture is getting out of date.

On November 18, 2026, GitHub announced that Copilot can now connect to eight more outside services through GitHub-verified MCP servers, which GitHub calls connectors, in public preview: Box, Buildkite, Calendly, ClickUp, Mixpanel, Supabase, Webflow and Wix. They join more than 25 services Copilot could already connect to, such as Slack, Figma and Sentry.

GitHub's own examples show why this is useful. You can ask Copilot why your latest build failed and have it fix the code in the same session, or have it check your Supabase database schema before it writes a migration. Connect a service once, and you can use it across Copilot.

That last part is the one I'd slow down on.

Why this is a business question, not just a developer setting

Once an assistant is connected to your company's file storage, scheduling, task tracker, analytics or database, it is no longer only touching code. It may be able to search client documents, see who is meeting whom, read product data, or work with the database behind your app. Whether that is fine depends on details most owners never see: which account the connection uses, what that account is allowed to do, and who switched it on.

GitHub does put some guardrails in place. Connectors are available on all Copilot plans, but on Copilot Business and Copilot Enterprise an admin has to enable the "Editor preview features" policy first. If your company's MCP allowlist names connectors one by one instead of using a wildcard, an admin also has to add the new ones. Some services, such as Box, also need an admin to allow MCP access on their side.

Those are good controls. But an admin policy only helps if someone owns it, and a connection made with a developer's own broad account can reach far more than the task needs.

Five things I'd check before connecting anything

  1. Which account does each connection sign in as? A personal admin account and a limited service account are very different risks.
  2. Can it change things, or only read them? Reading a database schema is one thing. Running a migration against production is another.
  3. Who on the team can turn connectors on? Decide whether any developer can, or only an admin.
  4. What data sits behind each service? Client files in Box, customer records in a database and meeting details in Calendly deserve different answers.
  5. How do you turn it off? Know where to disconnect a service and revoke its access the day someone leaves or a project ends.

None of this needs a big policy document. A short list of approved connections, who approved each one, and what account it uses is enough to start.

Where the coding tools stand

In Agent Ready Rankings' coding agents category, these are the current overall scores out of 10:

  • Cursor: 8.75
  • Claude Code: 8.4
  • Windsurf: 8.05
  • GitHub Copilot: 7.4
  • Aider: 7.3
  • Continue: 6.7

The overall score doesn't tell you whether a particular connection is safe for your business. That still comes down to the questions above.

About the rankings

I built Agent Ready Rankings, so treat the scores as one informed opinion rather than a verdict. Browsing is free at https://rankings.leffke.com/categories/coding-agents, and the full scoring breakdown is in Pro for $10/month.

Would you let your team's coding assistant connect to your company's files or database? I'd like to hear how you decide.