Share what Claude Code builds, and let it fix the feedback
Claude Code is good at building a page and not much use at getting five people to look at it. Ollav gives it a link to hand out. Reviewers pin comments on the page itself, and Claude reads them, fixes what they point at and publishes the next version to the same link.
Who this is for
You use Claude Code to produce HTML: a landing page, a dashboard mock, a report, a single-file React build. You want teammates or clients to react to the real page, not a screenshot, without you deploying anything.
How the loop works
1. Connect once
Create an API key at https://app.ollav.app/keys, then add the MCP connector:
claude mcp add --transport http ollav https://app.ollav.app/mcp \
--header "Authorization: Bearer $OLLAV_API_KEY"2. Publish
Ask Claude Code: "Publish prototype.html to Ollav and give me the link." It uploads one self-contained HTML file (up to 15 MB) with share_html and returns a link. No deploy step. The document is private to your workspace by default and not indexed by search engines.
3. Review
Reviewers sign in, click the element they mean and pin a comment, or react with an emoji. Workspace members and invited guests can comment, and guests are free.
4. Close the loop
Ask Claude Code: "Read the review, fix the open comments, resolve them and publish a new version." It reads the threads with get_review, replies and resolves with add_comment and resolve_comment, then ships publish_new_version. The link doesn't change and earlier comments stay.
The agent feedback loop guide goes through the prompts in more detail, and the MCP docs list every tool.
After you sign up, a short first-run checklist walks you through the connection.
When something else fits better
- You just need the file online. A static host such as Netlify Drop or GitHub Pages is simpler and has no review features. See Ollav vs. static hosting.
- It's a multi-file app or needs a server. Ollav hosts one self-contained file with no server code. Deploy it normally and use a feedback widget on the deployed site.
- Reviewers must comment without signing in. Ollav reviewers sign in, which keeps every comment attributable to a person. If you need fully anonymous comments, dot. allows them.
- You are reviewing a live, running site. Ollav reviews the file you upload, not your production URL. BugHerd is built for that.
FAQ
How do I connect Claude Code to Ollav?
Create an API key at https://app.ollav.app/keys, export it as OLLAV_API_KEY and run the claude mcp add command shown above. claude mcp list should then show ollav as connected. Other clients are covered in the agent quickstart.
What can Claude Code do with Ollav?
Publish an HTML file and get a link, read open comments as Markdown, reply to a comment, resolve a thread and publish a new version at the same link. The same actions are available over the REST API.
Does the link change when Claude publishes a new version?
No. Every version lives at the same link, and the comments from earlier rounds are kept, so reviewers don't have to find the new URL.
Can anyone with the link comment?
Reviewers sign in first, either as workspace members or as invited guests. Guests are free. A document is private to your workspace by default and isn't indexed by search engines.
What can I upload?
One self-contained HTML file up to 15 MB. Inline the CSS, JavaScript and images, or load them from absolute https:// URLs. A single-file build from a bundler works too. Relative asset paths and server-side code don't.
What does it cost?
There is a Free tier and you don't need a card to start. Pro is EUR 12 a month and Team is EUR 10 per seat a month with a minimum of 3 seats, both excluding VAT. See the pricing page.
Try it
Sign up on the Free tier, no card needed, and ask Claude Code to publish its next page.