How do I share what Cursor's agent wrote without screenshotting the sidebar?

Give Cursor the Comma MCP server and a rule that says where finished work goes. The agent publishes the HTML itself and hands you back a URL, so the artifact leaves the editor as a page rather than as a screenshot of a chat pane.

Why not commit the agent's report into the repo?

You can, and for anything that belongs in version control you should. But a migration plan, an audit, or a dependency review that goes into a repo is only readable by people who clone the repo and render the markdown — which excludes the PM, the security reviewer and the customer who asked the question.

Does this need Cursor's agent mode?

It needs MCP, which Cursor reads from .cursor/mcp.json in the project or from your global settings. Once the server is attached, the publish is a tool call the agent can make on its own; see the Cursor setup guide.

Can the agent update the same report on the next run?

Yes. Have it keep the report id — in the project rule, in a scratch file, wherever your team already keeps that kind of state — and update it. The URL stays the same and each run appends a revision, so a bookmark keeps working.

Can teammates reply where the agent can see it?

Yes. Comments anchor to the paragraph or code block they are about, and the agent can read them back through the same MCP server and act on them — see letting an agent respond to comments.

Share what Cursor's agent produced

You asked Cursor to audit the dependency tree, or plan the migration, or work out why the flaky test is flaky. It did a good job. The result is a long, well-structured answer in a sidebar in your editor, on your laptop.

Everything you do next is a downgrade. Screenshotting the pane loses half of it and all of the tables. Copy-pasting into Slack loses the structure and buries it above the next forty messages. Committing it to the repo means the only people who can read it are the people who clone the repo — not the PM who asked, not the security reviewer who needs to sign off, not the customer.

Let the agent publish it

Attach Comma's MCP server to Cursor (see Cursor setup — it is a few lines in .cursor/mcp.json and a scoped token), then tell the agent where finished work goes. A project rule in .cursor/rules is the right place, because it applies to every conversation without you having to remember:

When you finish a piece of work that a human needs to read — an audit, a
migration plan, a postmortem, a benchmark comparison — publish it to Comma as
an HTML report and reply with the URL. Do not paste the full text into chat.
Keep the report id in `.cursor/comma-report-id` and update that report on
later runs rather than creating a new one.

From then on, "audit our dependencies" ends with a link instead of a wall of text. The agent writes the HTML, calls the publish tool, and returns a URL you can paste into the thread where the question was originally asked.

Why a report and not a chat message

It renders. Tables stay tables, charts stay charts, the diff stays aligned. The markup is stored verbatim and served inside a sandboxed iframe, so what you saw is what your reader sees.
Readers need nothing. No Cursor licence, no editor, no repo access. The audience for "here is why the migration is risky" is rarely the set of people with a checkout.
Comments land on the sentence. Someone highlights step 4 of the plan and says "we can't do that until the vendor contract renews", and the thread stays pinned there — see commenting on HTML.
The agent can read the replies. Through the same MCP server, so the next prompt can be "address the open comments" rather than you relaying them by hand — see letting an agent respond to comments.
Re-running keeps the URL. Update the same report and the link in the thread still resolves, with the previous version in the revision history.

Worth knowing

Scope the token down. reports:write is enough for publishing; a token that can also read every report is a bigger blast radius than an editor plugin needs. See scoped tokens for AI agents.
HTML body: 5 MB, which is a great deal of prose and a little bit of base64.
Nothing re-executes. A published report is a snapshot. If you want it to refresh on its own, hand the job to a routine.

Try it

Comma is free — unlimited reports, unlimited commenters, unlimited revision history.

Attach the MCP server →

Share what Cursor's agent produced

Let the agent publish it

Why a report and not a chat message

Worth knowing

Try it

Related