How do I share a deep research report from an AI tool?

Export or copy it as HTML and publish it to Comma — by hand, or from the agent via the REST API or MCP server with a scoped comma_sk_ token. You get a URL where the full report renders, sources and all, and where reviewers can challenge individual claims in place.

Why does a research report need anchored comments specifically?

Because the failure mode of AI research is a confident sentence with a weak source, and that is a per-sentence problem. A general 'looks off to me' does not locate it. Highlighting the claim and pinning 'this cites a 2019 press release, not the filing' does.

How do I keep a research report current?

Have the agent re-run on a schedule and PATCH the same report id. The URL is stable, each run appends a revision, and diffing two revisions shows what changed in the findings — which is far more useful than re-reading 4,000 words.

Can the agent read the reviewers' objections?

Yes. Comma's MCP server exposes the comment thread as typed tools, so the agent can list unresolved comments on its own report, address them, and reply — a closed loop rather than a one-way publish.

Is the report private?

Reports are private by default, with access set per report — private, team-visible, domain-gated, or shared with named reviewers.

Share a deep research report

Deep research modes produce the longest artifacts any model writes: twenty sections, forty citations, an executive summary that is itself a page. They are also the least reviewed. The output arrives as a wall of text in a chat window, the recipient skims the summary, and nobody checks whether footnote 23 actually supports the sentence pointing at it.

That is not a model problem. It is a venue problem: a chat transcript has no way to mark one sentence as suspect.

Publish it as a document

Export the research as HTML — or have the agent post it directly:

curl -fsS -X POST https://commareports.com/api/v1/reports \
  -H "Authorization: Bearer $COMMA_API_TOKEN" \
  -H "Content-Type: application/json" \
  -d "$(jq -n --rawfile html research.html \
        '{title: "Market landscape — payments infra", html: $html}')"

Agents that speak MCP can do the same thing as a tool call, with the same scoped token — see where should my agent post and scoped tokens for AI agents.

Review the claims, not the report

The unit of review for research is the claim. On a published report, each one can be marked:

Challenge a sentence in place. "Source is a vendor blog, not the filing" pins to the sentence. See commenting on HTML.
Mark what is verified. Resolve the threads that check out, leave the rest open, and the open threads are the remaining work.
Let the agent answer. It can read unresolved comments back over MCP, fix what it got wrong, and reply — see let an agent respond to comments.

Keep it alive

Research goes stale fastest exactly where it is most useful — competitors, pricing, regulation. A routine re-runs the research on a cron and PATCHes the same report, so the link a stakeholder bookmarked in March still shows current findings in July, with every intermediate version in the revision history.

Limits

HTML body: 5 MB — long-form research fits comfortably.
Scripts run, sandboxed: allow-scripts, no allow-same-origin.
60 requests/minute per token.

Try it

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

Publish a research report →