You run a shop on Rakuten Ichiba and want Claude to help with your store data: analyse orders, check stock, review a campaign. The first question is how to get data out of RMS, Rakuten's merchant back office, and into Claude. This guide covers the three routes that work as of November 2026, with setup, cost and the data-protection points for each, plus a test in which Claude Code analysed a synthetic order CSV.
Short answer: as of November 9, 2026, Rakuten offers no official MCP server for Ichiba merchants, and Rakuten is not in Claude's connector directory. You have three practical options. One: download CSV files from RMS, delete the personal-data columns and give the files to Claude. Two: use the merchant API, RMS WEB SERVICE, through an MCP server you build, a community one, or the Orkas connector. Three: skip Claude and use Rakuten's own assistant inside RMS, Rakuten AI for RMS. For most shops, CSV is the safest place to start.
RMS is available only in Japanese. Menu names below are English descriptions; to see the exact Japanese labels, switch this page to 日本語.
Rakuten using Claude is not the same as merchants connecting Claude
Search for “Rakuten Claude” and most results are about Rakuten Group’s own use of Claude. In Anthropic’s case study, Rakuten’s engineers use Claude Code and cut the time to ship a new feature from 24 days to 5. Rakuten Mobile’s business service, Rakuten AI for Business, has also added Claude-powered features.
Those are stories about Rakuten the company. They are not a feature that lets Ichiba merchants connect RMS to Claude. From a merchant’s point of view, there is still no official link between Claude and RMS, so you choose the route yourself.
Three ways to use RMS data with Claude
| Option | What it can do | Setup | Watch out for |
|---|---|---|---|
| 1. Give Claude CSV files | Summaries of orders, traffic and reviews; campaign reviews; improvement ideas | Order CSVs need a separate application for the data download service | Delete personal-data columns first. Every download is manual |
| 2. RMS API plus an MCP server | Read products, stock and orders; update stock (depends on the tool) | Apply for the WEB API service and register an app. Renew the key regularly | No official MCP. Be careful with tools that can write |
| 3. Rakuten AI for RMS | Drafts of product descriptions, image edits, inquiry and review replies; data analysis | Built into RMS | Works only inside RMS. It is not Claude |
Options 1 and 2 need a Claude plan, paid separately. On November 9, 2026, Claude’s pricing page listed Free at $0, Pro at $20 a month, Max at $100 a month and Ultra at $200 a month, before tax.
Option 1: give Claude CSV files from RMS
No API setup is needed, so you can start today. Each run takes some manual work, but you decide exactly what the AI sees.
1. Download the CSV files in RMS
- Orders: export them with the data download function in order management, under the orders and payments menu. You first have to apply for the data download service, and each download asks for the user name and password you set when applying. A full-column export includes buyers’ personal data.
- Traffic: in the data analysis menu, the access and traffic analysis page exports a CSV for each product page.
- Reviews: the review check tool in the reviews menu exports a CSV that includes order numbers.
Menu names and locations change from time to time. If you cannot find one, check the manual inside RMS. Bulk editing of product data by CSV (CSV商品一括編集) is a paid option at ¥10,000 a month before tax (Rakuten’s service page, in Japanese).
2. Delete the personal-data columns
Sales and product analysis does not need buyers’ names, addresses, phone numbers or email addresses. Delete those columns in Excel or a similar tool, save the file and give that copy to Claude. If you will not match rows back to orders, remove the order numbers too.
Rakuten itself advises merchants to check beforehand how an outside AI service processes and stores the data they enter (Rakuten’s guide, in Japanese). On Claude’s individual plans (Free, Pro, Max and Ultra), you choose in the privacy settings whether your chats can be used to train models. Team, Enterprise and the API are not used for training by default.
3. Ask Claude
Attach the CSV to a chat and say exactly what you want back. The request below is an English translation of the Japanese prompt we used in the test that follows.
sample-orders-synthetic.csv contains order data from a Rakuten Ichiba shop (synthetic practice data; personal-data columns removed). Using only this CSV, summarise:
1) Sales by month (excluding cancellations) and the number of orders
2) The top five products by sales
3) The cancellation rate
4) The coupon campaign from September 15 to 21 compared with the week before and the week after (orders, sales, coupon discount and average order value)
5) Anything odd in the data, such as duplicates or negative amounts. If you find any, say how you handled them
Calculate every number from the CSV and do not fill gaps with guesses. Finally, suggest three things to try next month, each backed by a number.
If Japanese text in the CSV comes out garbled, name the encoding, for example “read it as Shift_JIS”. Ask Claude to show how it calculated each number, and check at least one total against the RMS screen.
Test: Claude Code analyses a synthetic order CSV
These results come from synthetic practice data. The file holds 631 order rows from August to October 2026 for a fictional tea shop, not data from a real store. The column names are simplified, so they do not match a real RMS export. It never had personal-data columns. We planted one exact duplicate order and one order with a negative total, and set a coupon campaign for September 15 to 21.
On November 9, 2026, we gave Claude Code 2.1.311 the Japanese original of the prompt above, unchanged, allowing it only to read files and run Python. It answered in about six minutes:
- It found both bad rows (synthetic data). It merged the duplicate order into one row. It excluded the negative row from the totals, giving reasons such as an out-of-sequence order number and a ¥1,000 coupon larger than the item price, and noted that if the order is real it should be checked in RMS.
- Monthly sales (synthetic data): August ¥411,490 from 168 orders, September ¥511,920 from 238, October ¥471,810 from 197. The cancellation rate was 4.13%.
- Campaign review (synthetic data): 39, 91 and 43 orders in the week before, the campaign week and the week after. Orders more than doubled, but it pointed out that the average order value in the campaign week, ¥2,084, was below the ¥2,372 average of weeks without a campaign. It suggested trying a coupon with a minimum spend next time.
- It kept outside knowledge separate. When it mentioned the September holidays, it labelled that as information from outside the CSV. It also said repeat rate and conversion rate could not be calculated, because the CSV has no customer IDs or traffic.
We recalculated the same CSV with a separate script, handling the two rows the same way (duplicate merged, negative row excluded). Monthly sales and order counts, product rankings, the cancellation rate and the campaign-week comparison all matched.
Excluding the negative row was Claude’s decision. In a real shop it could be a refund or a manual adjustment, and excluding it may be wrong. Always have a person check how unusual rows were handled. The run would have cost about $1.30 at API prices; in practice it used the allowance of a Claude subscription.
Option 2: connect through the RMS API (RMS WEB SERVICE)
To read orders or stock without downloading files each time, use the merchant API, RMS WEB SERVICE. It is separate from Rakuten Web Service, the API for shoppers and affiliates, which has no order or inventory APIs (Rakuten Web Service FAQ, in Japanese).
Get the keys
- In RMS, open the WEB API service from the information and services for merchants section, accept the terms and apply.
- Register an app and note its serviceSecret and licenseKey.
- Turn on each API you need (products, inventory, orders and so on) one by one. Enable only what you use.
The API documentation is open only to merchants logged in to RMS. The licenseKey expires, and most integration vendors say it must be renewed about every 90 days, or three months (for example, this STOCKCREW guide, in Japanese). When it expires the connection stops, so put the renewal date in your calendar. Vendors also announced that from June 2025, to use a tool from a developer registered with Rakuten, the merchant approves it in RMS, in the WEB API service access permission settings.
Ways to connect it to Claude
- Build your own MCP server: merchants have published write-ups, on note and elsewhere, of MCP servers they built for the RMS API with Claude Code. If you can read the documentation, you can build a small server that calls only the APIs you need.
- Use an unofficial MCP server from GitHub or elsewhere: a few RMS MCP servers are public. All are unofficial, built by individuals or companies, with few users so far, and some can write. Before using one, read its code to see which APIs it calls.
- Use the Orkas “Rakuten RMS” connector: covered in Using Orkas below.
The free server named Rakuten Remote MCP, from N-3 LLC, uses the public product search, ranking and travel search APIs. It does not connect to a shop’s RMS data (N-3’s announcement, in Japanese).
Option 3: Rakuten AI for RMS (RMS AI Assistant)
RMS has Rakuten’s own AI features. They launched in March 2024 as the RMS AI Assistant beta (Rakuten’s announcement, in Japanese), and are now also presented as Rakuten AI for RMS. They help write product descriptions, edit product images, draft replies to inquiries and reviews, and explain the R-Karte shop report. In April 2026 Rakuten added a data analysis agent that answers questions about sales and traffic in a chat, free for merchants on a store plan, according to press reports (in Japanese).
Because it all happens inside RMS, you never export a CSV to an outside service. On the other hand, outside AI tools cannot call it, and it is not built for analysis that combines your own CSV files or other marketplaces’ data, or for steps you define yourself. A realistic split: Rakuten AI for RMS for product pages and daily replies, Claude for analysis across several data sources.
Using Orkas
Orkas is a desktop app (macOS and Windows) that runs a team of AI agents. It can work with RMS data either from CSV files or through a connector.
Attach a CSV and ask
Attach a CSV exported from RMS (with the personal-data columns deleted) to a chat and ask as in the prompt above. If you have connected Claude Code to Orkas, you can hand the job to Claude Code. Without a Claude plan, you can try it with the built-in Orkas model and the 500 credits new users receive. Orkas’s built-in agents can read CSV files in Shift_JIS and other encodings.
The “Rakuten RMS” connector
Open Connectors in Orkas and choose the E-commerce filter to find Rakuten RMS. If you search, type “Rakuten” or “RMS”. After you enter the serviceSecret and licenseKey issued in RMS, the keys are stored encrypted on your computer, and your computer connects to Rakuten’s API directly, without going through Orkas’s servers.
By design, the connector can read the product list and product details, stock for each SKU, one day’s list of order numbers and order details, and change the stock count of an existing SKU. Nothing else: it cannot change prices, product pages or order status. Before a stock change, the default settings show a confirmation dialog with Allow once, Allow for this task and Don’t run. If you choose Allow for this task, it will not ask again within that task, and in Trusted mode no dialog appears.
To be clear, this connector has only been tested against mock data. We have not yet confirmed that it works with a real Rakuten shop. Order details include buyer information, which is passed to the AI model as is. There is no warning before the licenseKey expires; when it does, you reconnect. If you use it, start by reading products only, and read orders only when you need them.
Personal data and Rakuten’s rules
- Buyer data is entrusted to your shop for the transaction. Rakuten’s privacy policy (in Japanese) says buyer information is provided to shops for the transaction and managed by the shop. As a rule, do not give an AI columns your analysis does not need.
- We could not find public information on how the merchant terms treat outside AI. The merchant terms and guidelines are visible only after logging in to RMS. Check with 店舗運営Navi, Rakuten’s merchant help site, or your Rakuten contact.
- Avoid letting a browser-automation AI operate the RMS screens. We found no published rule saying whether Rakuten allows automated operation of its screens. To get data out, use the API or CSV.
- Have a person check what the AI writes. Rakuten also advises always checking AI-generated text and images by eye, with care for wording covered by Japan’s Pharmaceutical and Medical Device Act and its law against misleading representations.
Which option to choose
| If you want to | Use |
|---|---|
| Review sales or a campaign once a month | Option 1: give Claude CSV files |
| Draft product descriptions or replies to inquiries | Option 3: Rakuten AI for RMS |
| Have an AI check stock or orders every day | Option 2: RMS API plus MCP, if you can manage key renewal and permissions |
| Combine data from other marketplaces or ads | Option 1: give each service’s CSV files to Claude or Orkas together |
FAQ
Does Rakuten have an official MCP server or Claude connector?
We found none as of November 9, 2026. Rakuten has not announced an official MCP server for Ichiba merchants, and Rakuten is not in Claude’s connector directory. The MCP servers published for RMS are unofficial, built by individuals or companies.
Who can use the RMS API?
Rakuten Ichiba merchants, who apply through the WEB API service under the merchant information and services section of RMS. The documentation is also open only to merchants logged in to RMS. Rakuten Web Service, for shoppers and affiliates, has no order or inventory APIs.
Can I give my order data to Claude as is?
We do not recommend it. Sales and product analysis does not need buyers’ names, addresses, phone numbers or email addresses, so delete those columns first. Also check Rakuten’s terms and your shop’s own privacy policy on whether you may share the data with an outside AI.
When should I use Rakuten AI for RMS instead of Claude?
For work that stays inside RMS, such as drafting product descriptions or replies to inquiries, Rakuten AI for RMS is the easy choice. Claude suits analysis that combines several CSV files or other marketplaces’ data, and calculations that follow steps you set.
How much does Claude cost?
On November 9, 2026, Claude’s pricing page listed Free at $0, Pro at $20 a month, Max at $100 a month and Ultra at $200 a month, before tax. Since April 1, 2026, customers in Japan are also charged 10% consumption tax.