Attribura

Built for agents

Your agent already has the keys.

Attribura ships a hosted MCP server. One command points Claude at your account, and from then on the panel is something you ask rather than something you open.

terminal
$ claude mcp add --transport http attribura https://api.attribura.com/mcp -H "Authorization: Bearer atb_…"

Read

get_overviewget_analyticsget_revenue_breakdownget_performancelist_contentget_reputation

Keywords

suggest_keywordstrack_keywordslist_tracked_keywordsget_keyword_historykeyword_brief

Write

update_listingreserve_keyworddraft_replysend_replypush_creativepublish_creative

read-only by default · scoped tokens · one command

A dashboard is a place you have to go.

Every tool you own is a tab, and every tab is a small tax on wanting to know something. So you stop wanting to know. The question you would have asked on a Tuesday — did that reel actually sell anything, is that keyword still climbing — quietly turns into a thing you will check later, and later never has a date on it.

Agents change what a tool has to be. If Claude can read your numbers and act on them, the panel stops being somewhere you visit and becomes something that answers. That only works if the tools are real, scoped, and honest about which of their numbers are estimates — which is exactly what Attribura was already built to be.

How you connect it.

01

Mint a token

Scoped, and read-only unless you say otherwise. You choose what an agent is allowed to touch before it touches anything.

02

Run one command

One line points Claude Code, Claude Desktop or any MCP client at the hosted server. There is nothing to install and nothing to run yourself.

03

Just ask

Your brand, your personas, your keywords, your posts and your revenue are all readable, and the write tools do the work you would have clicked through.

What you can just ask for.

“How did last month go?”

Installs, proceeds and which posts brought them in — read out of your real account, not summarised from a screenshot.

“Find keywords worth taking.”

Terms pulled off the apps outranking you, tracked, and the ranks read back a week later without you asking twice.

“Rewrite my subtitle for the US.”

Edits the live listing inside Apple's character limits, in any language you ship.

The specifics

Transport
Hosted HTTP MCP server at api.attribura.com/mcp. Nothing to install, nothing to keep running.
Auth
Scoped bearer tokens you mint and revoke. Read-only by default; write access is opt-in per token.
Reads
Overview, analytics, revenue and its breakdown, content, clicks, keywords and their history, the store listing, reputation, self-reports.
Writes
Track and release keywords, update the listing, reserve a comment keyword with its link, draft and send review replies, push and publish creatives.
Provenance survives
The tools return the same labels the screens do — Apple's number, our estimate, approximate — so an agent cannot quietly promote a guess to a fact.
Clients
Any MCP client. Claude Code and Claude Desktop are one command each.

Questions people ask.

What is an MCP server?

A standard way for an AI client to call a product's tools. Instead of pasting screenshots into a chat, the model reads your actual data and takes actual actions through a defined interface. Attribura hosts one, so there is nothing to install and nothing running on your machine.

Can Claude change things without asking me?

Only if you mint a token that lets it. Tokens are scoped and read-only by default, so the normal setup is an agent that can read everything and change nothing. Write access is a deliberate choice you make per token and can revoke at any time.

What can it actually do?

Read your overview, analytics, revenue and its breakdown, your content, your keyword ranks and history, your store listing and your reviews. With write scope it can track keywords, rewrite the listing, reserve a comment keyword and mint its link, draft and send a review reply, and push a finished creative into your library.

Does it work with anything other than Claude?

Yes. It is a standard MCP server over HTTP, so any MCP client can use it. Claude Code and Claude Desktop happen to be one command each, which is why the docs start there.

How does it avoid making things up?

The tools return the same provenance the screens carry — this figure is Apple's, this one is our estimate, this position is approximate. A model that is handed those labels repeats them. A model handed bare numbers invents a confidence nobody earned, which is the failure mode worth designing against.

Put your app in.

Sign in, connect one account, and watch the first front go live. It takes a few minutes, and the panel tells you what to do next.