# Share What opencode Produced — A Link, Not a Terminal Pane

Canonical: https://commareports.com/agents/share-opencode-output
Published: 2026-09-09

> opencode writes the audit, the plan, the postmortem into a TUI session on your machine. Attach the Comma MCP server and it publishes to a URL your team can open and comment on.

# Share what opencode produced

A terminal agent has the tightest loop and the worst distribution. opencode
runs where your code is, reads what it needs, and writes a genuinely useful
document — the dependency audit, the migration plan, the postmortem draft — into
a TUI pane on a machine nobody else has.

Everything you do with it next loses something. Redirecting to a file gives you
an HTML file, which is the same problem with extra steps: emailed, it opens
from a downloads folder with broken relative paths. Pasting into Slack strips
the tables. Committing it means only people with a checkout can read it.

## Let the agent publish it

Add Comma's [MCP server](/mcp) to the `mcp` block in `opencode.json` with a
[scoped token](/docs/api-tokens), then write the standing rule once in
`AGENTS.md` at the repo root:

```md
## Publishing finished work

When you finish something 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 print the whole document to the terminal.

The report id is in `.comma-report-id`. Update that report rather than creating
a new one.
```

`AGENTS.md` is the file most agents already read, so the same rule tends to
work across whichever tool a teammate happens to be using — see
[where should my agent post?](/agents/where-should-my-agent-post).

## Why a report and not a pane

- **It renders.** Tables stay tables, the diff stays aligned, a chart the agent
  emitted stays a chart. Markup is stored verbatim and served in a sandboxed
  iframe.
- **Readers need nothing** — no terminal, no repository access, no licence for
  the tool you happen to run.
- **Comments land on the sentence** they are about, and stay pinned there — see
  [commenting on HTML](/comment-on-html).
- **The agent reads the replies** through the same MCP server, so "address the
  open comments" is a prompt — see
  [letting an agent respond to comments](/agents/let-an-agent-respond-to-comments).
- **Reruns keep the URL.** One id, one link, a revision per run.

## Worth knowing

- **Scope the token to `reports:write`.** Add `comments:write` only for the
  reply half of the loop — see
  [scoped tokens for AI agents](/agents/scoped-tokens-for-ai-agents).
- **Rate limit: 60 requests/minute per token**, which a terminal agent will
  never notice and a parallel fan-out might.
- **HTML body: 5 MB.**
- **Nothing re-executes.** A published report is a snapshot; scheduled
  regeneration is a [routine](/docs/routines).

## Try it

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

**[Attach the MCP server →](/mcp)**

### Related

- [Share what Claude Code produced](/agents/share-claude-code-output) ·
  [Share what Aider produced](/agents/share-aider-output)
- [Share what Warp's agent produced](/agents/share-warp-output) ·
  [Give an agent feedback](/agents/give-an-agent-feedback)
- [MCP setup for every client](/mcp) · [The API](/docs/api)