Statory statistics inside Claude

What it can do

• Call the seven core analyses (frequency, descriptives, crosstab, t-test, ANOVA, correlation, regression) directly as tools
• Pick and run any of the 106 analyses that are open to the connector
• Get the same tables, the same APA sentence, and the same "how many cases" contract you see on the web
• Optionally receive chart specifications (JSON)
• Answers in English or Korean — it follows the language of the table you send, and you can set it with the language argument

What it does not do

• AI interpretation — there is no such tool here. It produces numbers and tables only.
• File upload or CSV parsing — that lives on the web app (/intake).
• Anonymous use — the first connection asks you to approve with a Statory account.

How to connect

Claude

1. Open Settings → Connectors (or Integrations).
2. Choose "Add custom connector".
3. Paste Statory's MCP address.
4. Press "Connect"; a browser window opens with the Statory sign-in page.
5. After signing in you see the approval screen. Check the client name and the return address, then press "Approve". If the return address is one we have not seen before, the screen says so.
6. You are returned to the client and the connection is done. Declining leaves nothing behind.

ChatGPT (not open yet)

1. Turn on Settings → Security and login → Developer mode.
2. Go to Plugins → Browse plugins.
3. Press the ＋ next to the search box.
4. Fill in name, URL and OAuth, then press Create.
5. Press Connect; the Statory sign-in and approval screens appear — the same ones Claude shows.
6. In a conversation, reach it through ＋ → More → Statory.

The 9 tools

All nine are read-only: nothing is deleted or modified, and nothing leaves Statory.

The options field of run_analysis

options is a free-form object: each engine accepts different keys. Its shape is the same as the options of the web request (/api/v2/analyze). The list of open analyses comes back from the list_analyses tool, and on the web each analysis shows the same keys in its engine-settings panel.

When it is absent the engine default applies; nothing is silently substituted.

Where the data goes

• The table you send goes to the engine only; it is not stored after the calculation.
• The response never carries the raw data back — only aggregated tables and the report sentence.
• One line is written to your analysis history (which analysis, which variables, how many cases). The data itself is not in that line.
• You can read that history at /history; deleting your account follows the account-deletion policy.
• No credits are charged for MCP runs.

Limits

• A table can hold up to 1,000,000 cells (rows × columns), with at most 1,000 columns.
• Each request body is capped at 12MB.
• Each account can run 60 analyses per hour.
• An analysis that runs past 120 seconds stops and says so.

Frequently asked

I cannot find "custom connector" in ChatGPT.

The ChatGPT connection is not open yet, so there is nothing to add there for now. Use Claude in the meantime. When it opens, the step is Settings → Security and login → Developer mode.

The connection ends with "registration failed".

The authorization server setting is not switched on yet. Please let us know.

The answer comes back in Korean and I want English.

If the column names or values in the table you sent contain Korean, the answer comes back in Korean. Tell the assistant "run it again with language set to en" and it will come back in English (and the other way round).

"Sign-in required" keeps appearing.

Usually the permission is still valid and only the token has expired. Statory now says "expired, please refresh" in the response, so the AI client fetches a new one by itself. If it still repeats, remove the connector and add it again — in that case the token was not issued by this server.

The consent screen says "the request could not be read".

If you reconnect a connector you have already approved, Statory sends you straight back without a consent screen — there is nothing left to press. This used to stall there and is now fixed. If you see "this request has already been handled", start the connection again from the AI client — a browser refresh or Back will not do it, because a connection request is spent once it is used.

It says the analysis "is not open".

That engine is not exposed over the connector. You can still use it on the web.

It says the table is too large.

There is a limit on how much can be sent at once. Split the rows, or send only the columns the analysis needs. The limits are listed on this page.

It says "up to N times per hour".

There is an hourly limit on runs. Try again in a little while.

I want to upload a file.

For now the connector takes tables (values) only. Upload files on the web instead.

I cannot see it in the ChatGPT app list.

The ChatGPT connection is not open yet, so it is not in that list. Claude is open today — the steps are on this page.