How do I share what Warp's agent produced without pasting a terminal block?

Attach Comma's MCP server in Warp's MCP settings and add a rule to WARP.md that says finished work gets published. The agent writes the HTML, calls the publish tool, and hands back a URL — so the artifact leaves the terminal as a page instead of as a screenshot of a block.

Warp already lets me share a terminal block. Why isn't that enough?

A shared block is a transcript of commands and their output — excellent for 'here is what I ran'. It is the wrong shape for 'here is the analysis', which has tables, sections and numbers a reader needs to scan rather than replay, and which someone needs to leave a comment on.

Where does the standing rule live?

WARP.md in the repository root, which Warp reads as project context for every conversation in that directory. Put the publish instruction there once and every task in the repo ends with a link rather than a wall of text.

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

Yes. Keep the report id in the repo — a line in WARP.md, a dotfile, wherever your team already keeps that kind of state — and have the agent PATCH it. The URL stays stable and each run appends a revision.

What scope does the token need?

reports:write to publish. Add comments:write only if you want the agent to reply in threads. A token that can also read every report in the account is a larger blast radius than a terminal needs.

Share what Warp's agent produced

Warp's agent is good at the jobs that end in a document rather than a diff: work out why the deploy stalled, read the slow query log and explain it, walk the migration and list what breaks. It does the work in your terminal and the result lands in a block — a very nice block, with syntax highlighting, on your laptop.

Then somebody asks you to send it to them, and every option is worse than what you are looking at. A screenshot of a scrollback pane loses the tables. Pasting it into Slack loses the structure and drops below the fold in an hour. Sharing the block sends a transcript of commands, when what the reader wanted was the conclusion.

Let the agent publish it

Attach Comma's MCP server in Warp's MCP settings with a scoped token, then put the standing instruction in WARP.md at the repo root — Warp reads it as project context for every conversation in that directory, so you never have to remember to ask:

## Publishing finished work

When you finish something a human needs to read — an incident timeline, a
query plan analysis, a migration checklist, a capacity review — publish it to
Comma as an HTML report and reply with the URL. Do not paste the full document
into the terminal.

The report id is in `.comma-report-id`. Update that report rather than
creating a new one, so the link in the runbook keeps working.

From then on "explain why last night's deploy stalled" ends with a URL you can paste into the incident channel, and the terminal stays a terminal.

Why a report and not a block

It renders. Tables stay tables, the timeline stays aligned, the chart the agent generated is still a chart. Your markup is stored verbatim and served in a sandboxed iframe, so the reader sees what you saw.
Readers need nothing. No Warp licence, no shell access, no SSH to the box the agent was working on. The audience for "here is why the deploy stalled" is mostly people without a terminal open.
Comments land on the sentence. Someone highlights the step where the timeout was raised and asks who approved that, and the thread stays pinned to that line — see commenting on HTML.
The agent can read the replies through the same MCP server, so the next prompt is "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 you pasted into the runbook still resolves, with the earlier version in the revision history.

Worth knowing

Scope the token down. reports:write is enough to publish. Tokens expire, revoke individually and rate-limit at 60 requests/minute — see scoped tokens for AI agents.
HTML body: 5 MB. That is a great deal of prose and a modest amount of base64. A profile or a trace bundle goes up as an attachment instead.
Nothing re-executes. A published report is a snapshot of the markup. If you want it refreshed on a schedule without anyone asking, that is a routine.
Secrets. An agent that just read your logs may have interesting strings in context. Publish to a private report first and widen the access setting once you have looked at it — see sharing & access control.

Try it

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

Attach the MCP server →

Share what Warp's agent produced

Let the agent publish it

Why a report and not a block

Worth knowing

Try it

Related