Cursor for Microsoft Teams: Start Cloud Agents with @Cursor

It connects Cursor's Cloud AgentsAgents that run in a Cursor-managed virtual machine, check out the repo, do the work and open a pull request, then shut down, with no load on your laptop. Press Enter for the full definition. to Microsoft Teams so you can kick off coding work without leaving the conversation. You mention @Cursor with a prompt, and an agent starts against the right repository. The integration supports personal chats, team channels, and group chats.

None of the work happens inside Teams. The agent runs as a Cloud Agent on Cursor's side, against a repository connected through your Git provider.

The value shows up in a narrow case, which is when the task has already been described in the channel. Someone reported a bug, two people worked out what was wrong, and the fix that is left is one sentence. Mentioning @Cursor there saves restating all of it somewhere else.

Installation starts from the Cursor integrations dashboard and finishes back in Cursor, where you connect a repository provider and confirm settings. These are Cursor's documented steps.

1. 1Go to Cursor integrations and click Connect next to Microsoft Teams (or open the Microsoft Marketplace listing).
2. 2Install the Cursor app in your Microsoft Teams workspace.
3. 3After installing, you are redirected back to Cursor to finalize setup.
4. 4Connect GitHub, GitLab, Azure DevOps, or Bitbucket if you have not connected a repository provider yet.
5. 5Enable usage-based pricing and confirm your privacy settings.
6. 6Start using Cloud AgentsAgents that run in a Cursor-managed virtual machine, check out the repo, do the work and open a pull request, then shut down, with no load on your laptop. Press Enter for the full definition. in Teams by mentioning @Cursor.

That order makes the Teams install sound like the first move in adopting Cloud AgentsAgents that run in a Cursor-managed virtual machine, check out the repo, do the work and open a pull request, then shut down, with no load on your laptop. Press Enter for the full definition., when it sits nearer the end of the job. An app can be installed in Teams while the Cursor side is unfinished, and an agent with no connected repository provider has nothing to work on. Cursor's advice is to get the environment configured before you run anything, so get a few runs working from Cloud Agent setup and add Teams after that.

Settle the privacy question before you ask anyone to install anything. Cloud AgentsAgents that run in a Cursor-managed virtual machine, check out the repo, do the work and open a pull request, then shut down, with no load on your laptop. Press Enter for the full definition. do not support Privacy ModeCursor's setting that guarantees code data is not used for training by Cursor or its model providers, and that an admin can enforce org-wide; data-retention terms are a separate, contractual layer. Press Enter for the full definition. (Legacy), so an organization on that setting has nothing here to pilot.

Team size decides where the list stalls. Two engineers, and one of them does all of it in a sitting. Bigger orgs tend to stall in the middle, because enabling usage-based pricing is a billing decision and whoever wants the integration is not always the person who can approve it.

Mention @Cursor and give your prompt; Cursor picks the repository and model automatically. To pin a specific repository or model, name it in the message or use inline options. The table below lists the documented commands.

Options can be written naturally or inline, and the bot parses them from anywhere in the message. Cursor documents both shapes, and the two lines below are its examples of each. They are not the same request: read them as two syntaxes, not as a before and after.

@Cursor with opus, fix the login bug in backend-api
@Cursor repo=acme/backend branch=dev model=opus Fix the login bug

Both forms work, so the choice is about who else reads the message. Natural phrasing stays readable to the humans in the channel. Inline repo= and model= are precise and a little ugly, and that trade is worth making when you point an agent at a repository you do not normally touch. In your usual repo, with dashboard defaults set, neither is worth typing.

The table above is a snapshot of the docs. @Cursor help returns the live list.

Two rows there carry the same description. unlink and disconnect both disconnect your Cursor account from Microsoft Teams. I read them as aliases. The docs do not confirm that, and either way it takes one person's link out and leaves the app in place for everyone else.

Cursor evaluates your message in a fixed order, so explicit options always win. Dashboard defaults are the fallback. Knowing the order helps when an agent picks an unexpected repo.

1. 1Explicit options: repo, branch, and model values written in your prompt.
2. 2Your message content: repository, model, or branch names mentioned in the prompt.
3. 3Recent agent activity: repositories you have used recently.
4. 4Default repository: the fallback when no match is found.

Explicit values sit at the top of that list, and they are the ones worth typing, since inference fails quietly. A missing repository name does not raise an error. At worst it produces an agent working in the wrong codebase, and you find out when a pull request appears somewhere you were not watching.

Step three is the one I would read twice. Recent agent activity means repositories you have used recently, so it resolves per person rather than per channel. Two people in the same conversation can send the same sentence and start agents on different repositories, and nothing in the message shows that it happened. On a team with one repository this never comes up. Once there are several repos and several teams in the same channel, naming the repo is worth making a habit.

Base branch runs on a shorter ladder, and the note below has the mechanics. Left blank it falls through to the repository's default branch, often main. A team that integrates on a dev line and never sets that default gets Teams-started agents branching from main whenever nobody names one, which the launch card does show if anyone checks it.

In a channel thread you can. Reply in the agent's thread with another @Cursor mention and it lands as a follow-up instruction. In personal chats and group chats you cannot, so you carry on from Cursor instead, through Open in Web or Open in Desktop on the card.

That asymmetry catches people out, because the natural way to try a new bot is privately. You message it, the first result looks good, and the second instruction has nowhere to go except out of Teams and into the web app. Which is a fine way to work, just not the thing you set out to test.

So run the trial in a channel thread even if a private chat feels safer. Start it where the follow-ups will have to happen.

When an agent starts, Teams shows a launch card with the selected repository, model, and branch, plus options to Open in Web, Open in Desktop, and Switch repository. On completion you get a notification with the result, and if the agent opened a pull request the card links to it. You can also add follow-ups, switch repository, or delete the agent from the card.

Agent summaries and diff images are the part to settle before this goes into a busy channel, since they can carry file paths and code snippets. The Cloud AgentsAgents that run in a Cursor-managed virtual machine, check out the repo, do the work and open a pull request, then shut down, with no load on your laptop. Press Enter for the full definition. settings carry two admin-only toggles for it, one for the summary itself and one that extends it to connected external channels. Your editor sidebar and a channel with the whole department in it are not the same audience, so the second one is what matters here.

Read the launch card first, then work outward. It names the repository, model, and branch, so most surprises are visible there before the agent gets far.

The app is installed once for the workspace, but the account link is per person, which is what messageTeamMembers covers: Cursor sends direct messages for setup, account linking, and notifications. When one person's mention behaves differently from everyone else's, check their direct message from the bot before any dashboard setting.

If the repository is what went wrong, use Switch repository on the card. It relaunches the same request against a different repository.