Documentation
AI agents (MCP)
Your clients pin it. Your agent fixes it. Connect an AI tool to VibeVaults and it can read the feedback you can see, with the pin, the page, the console and the screenshot, and, only if you allow it, reply and change status.
What this is
MCP (Model Context Protocol) is the standard way AI tools connect to other products. VibeVaults runs an MCP server, so a tool such as Claude Code, Claude Desktop or Cursor can look up a report, see where on the page it was pinned, read the thread, and start fixing the site, without you copying anything out of the dashboard.
It is available on every plan and during the trial. Only owners and members can connect a tool. Clients and guests have no account, so they cannot.
Connect a tool
The server address is https://www.vibe-vaults.com/api/mcp. You never create or paste a key: the tool opens a VibeVaults page, you sign in if needed, choose what the connection may do, and approve.
Claude Code
claude mcp add --scope user --transport http vibevaults https://www.vibe-vaults.com/api/mcpThen run /mcp inside Claude Code and choose vibevaults to sign in.
Claude Desktop
Open Settings, then Connectors, choose Add custom connector, and paste the server address.
Cursor
Add this to your mcp.json:
{ "mcpServers": { "vibevaults": { "url": "https://www.vibe-vaults.com/api/mcp" } } }The same snippets, with copy buttons, are on your Account page.
What you decide when you approve
- Which workspaces. All of yours, or just one. A tool limited to one workspace cannot see the others at all.
- Read only, or read and write. Every new connection is read only. Tick Allow replying and changing status to let it reply to clients and move reports between statuses.
- AI-assisted marker. On by default. Replies the tool sends are labelled as written with AI help. Turn it off only if you review every reply before it goes out.
You can change all three later, or disconnect, from the AI agents (MCP) card on your Account page. Disconnecting takes effect on the tool's very next request.
What an agent can do
| Tool | What it does | Needs write |
|---|---|---|
list_projects | Your workspaces and projects, with how many reports are open, in progress and in review | No |
list_feedback | Reports for a project, filterable by status, by page, and by whether they are pinned | No |
get_feedback | One report in full: the pin and how reliable it is, browser details, console logs, failed requests, the thread, and up to three screenshots | No |
reply_to_feedback | Posts a reply in the thread, as you | Yes |
update_feedback_status | Sets a report to open, in progress, in review or completed | Yes |
It cannot delete anything, create reports, change members, projects or billing, or upload files. You can paste a dashboard link to a report straight into the tool and it will find it.
What protects your account
- It sees what you see. The connection acts as you, under the same access rules as the dashboard. It cannot read a workspace you are not in.
- Your plan decides, not the tool's. If the workspace owner's trial or subscription lapses, connected tools stop working for everyone in that workspace, reads included, exactly as the dashboard does. Disconnecting always keeps working.
- Client text is treated as untrusted. Anything a client or their web page wrote, such as feedback, replies and console output, is handed to the agent inside marked blocks with a notice that it is content to analyse and not instructions. That limits, but cannot fully remove, the risk of a hostile page or a sneaky comment steering an agent. Keep write access off for tools you do not fully trust.
- Limits. Each tool is limited to 60 calls a minute per person.
Replies sent by an agent
A reply posted through a tool is an ordinary reply from you: it appears in the thread, in the widget, and in the notification email your client receives. VibeVaults cannot tell whether you read it before your tool sent it, which is why the AI-assisted marker exists and why write access is off until you ask for it. You remain responsible for what is sent in your name.
Where the data goes
When a connected tool reads a report, that content is sent to the tool and to the AI provider behind it, at your instruction. That provider acts for you. It is not a VibeVaults sub-processor and we are not a party to your arrangement with it. Check that its terms suit what your clients have agreed to, and mention it in your own privacy notice where it applies. What a report contains in the first place is described in What the widget records.
If something does not work
- “This connection is read-only”: switch on Allow replying and changing status for that connection on your Account page.
- A workspace shows as paused: the owner's trial or subscription has lapsed. Nothing is deleted; it works again once the owner renews.
- The tool says it is not authorised: the connection was disconnected. Connect it again from the tool.
- The approval page says the request expired: start the connection again from the tool. The page only works when opened by the tool.
Something missing from this page? Email support@vibe-vaults.com and we will add it.