# Share Gemini CLI Output — Publish the Answer, Don't Paste It

Canonical: https://commareports.com/agents/share-gemini-cli-output
Published: 2026-09-05

> Gemini CLI's long answers live in a terminal nobody else has. Attach an MCP server, write the rule into GEMINI.md, and the agent publishes an HTML report and replies with a link.

# Share what Gemini CLI wrote

Gemini CLI is generous with output. Ask it to compare three approaches to a
schema migration and you get something genuinely worth circulating — laid out
in a terminal, on your machine, in a buffer that scrolls away.

The usual next steps all cost something. A screenshot captures a third of it.
A paste into Slack flattens the comparison table into ragged text. `> report.md`
produces a file whose only reader is whoever you email it to, and email is
where documents go to be ignored.

## Attach the server, then write it down

Comma is a remote MCP server, so setup is an `httpUrl` entry — the full block
is in [Gemini CLI setup](/mcp/gemini-cli), and the token comes from
[API tokens](/docs/api-tokens):

```json
{
  "mcpServers": {
    "comma": {
      "httpUrl": "https://commareports.com/api/mcp",
      "headers": { "Authorization": "Bearer comma_sk_..." }
    }
  }
}
```

Then put the standing instruction in `GEMINI.md`, which the agent loads on
every run:

```md
Finished work that a human needs to read — comparisons, audits, migration
plans, weekly digests — gets published to Comma as an HTML report. Reply with
the URL only. Reuse the report id in `.gemini/comma-report-id` so the link
stays stable across runs.
```

## Non-interactive runs benefit most

An interactive session has you watching it. A cron entry does not:

```bash
gemini -p "Summarise this week's API latency regressions and publish the \
report to Comma. Reply with only the URL."
```

One line of output, one URL, one page the on-call engineer can actually open.
If the run belongs in a pipeline see [CI reports](/ci); if you would rather not
own the scheduler at all, see [routines](/docs/routines).

## What the report gives you that the terminal does not

- **Rendering that survives.** Tables, code blocks and embedded charts arrive
  as the agent wrote them, in a sandboxed iframe.
- **A stable address.** The link you pasted last Tuesday still resolves, and
  the version behind it is in the revision history.
- **Anchored replies.** A reviewer highlights the one number they doubt rather
  than describing which paragraph they mean — see
  [commenting on HTML](/comment-on-html).
- **A read path back.** The agent can pull the open threads through the same
  server and act on them — see
  [letting an agent respond to comments](/agents/let-an-agent-respond-to-comments).

## Worth knowing

- **Scope the token.** `reports:write` for publishing; nothing else is needed
  to make this work. See [scoped tokens](/agents/scoped-tokens-for-ai-agents).
- **5 MB of HTML** per report body.
- **A report is a snapshot.** No re-execution, no live query, no surprise cost
  when forty people open it.

## Try it

Free, with unlimited reports, commenters and revisions.

**[Set up Gemini CLI →](/mcp/gemini-cli)**

### Related

- [Where should my agent post?](/agents/where-should-my-agent-post)
- [Share Claude Code output](/agents/share-claude-code-output) ·
  [Share Codex output](/agents/share-codex-output)
- [MCP setup for every client](/mcp) · [The API](/docs/api)
