Connect Your AI

How It Works

Set up in your client

Cursor and VS Code add the server in one click. Claude Code and Codex take one command, which the button copies. Claude on the web or the desktop takes one paste.

The commands in full

Run either once. claude mcp list and codex mcp list should then show the server as connected.

Other clients

Any client that supports remote MCP servers over Streamable HTTP can use the same URL. The server is stateless and requires no authentication.

Once connected, the full manual covers every lookup, how to read the trust labels in what comes back, and what to do when something looks wrong.

Example Lookup live from this server

What your assistant sends and gets back when you ask a plain question. The figures below were fetched from the server when you opened this page.

The assistant writes its answer from that result, and quotes the citation link so you can check the figure on the page it came from.

Coverage 82 datasets

The assistant sees every dataset through the same twelve lookups, so you never have to name one yourself. The Manual lists each dataset, and the datasets directory is the authoritative list.

Bars run from the first period held to the latest. Company filings accumulate filing by filing and have no single start year.

Example Questions

• What is Japan's headline and core inflation right now, and which groups are driving it?
• How broad is inflation across individual items — what share is rising more than 2%?
• How much JGB did the Bank of Japan buy last month, net of redemptions?
• Where is the 10-year JGB yield against a year ago?
• Which visitor markets grew fastest last month, and how full were Kyoto's hotels?
• Who holds Mitsubishi UFJ as a policy shareholding, and what reason do the filers state?
• Which listed companies have the oldest boards? Which directors got the lowest approval at the AGM?
• Show me Toyota's financial statements as filed, then its ROE against its Core30 peers.
• Which companies have the largest unspent buyback authorisations?
• What did last month's CPI look like as of the day before the latest release?
• Which Apple insiders sold shares in the last year, and how much?
• How has the 10-year Treasury moved against the 10-year JGB this year?
• What is Hong Kong's inflation rate, and how does it compare with Japan's?

Lookups 12, read-only

Each lookup is a thin wrapper over this site's public API, so the assistant can only see numbers the site already publishes, from the same accepted release. Your assistant picks the lookup; you never call one by name.

List datasetslist_datasets

Every dataset: what it is, its shape, and which lookups apply.

Describe a datasetdescribe_dataset

Its source, every measure with its trust label and formula, and what this server holds.

Searchsearch

A company by name, securities code or ticker, or a series by name or code, across all datasets.

Get a companyget_company

One dataset's full view, or a profile across every dataset that knows the company.

Headline readingget_overview

Where a dataset stands now: the figures its own page leads with.

Get series historyget_series

Published values or calculated rates, as of any date.

A dataset's own cutget_breakdown

Contributions to inflation, breadth of price rises, the whole yield curve, one product's trade with every partner.

Ranked screenscreen

The dataset's own screens: oldest boards, largest unspent buybacks, cheapest land, lowest director approval, highest ROE.

List peer groupslist_cohorts

The TOPIX scale bands, JPX market segments and the industry classifications a company can be compared within.

Compare within a peer groupcompare_cohort

One metric across one peer group, ranked, with the group's own median and quartiles.

Releases and revisionsget_vintages

Every release of a dataset, and how one figure was revised from release to release.

Check a published figurecheck_claim

A cross-shareholding number from the press, checked against what the filings support.

The dataset cards are also readable as MCP resources. The older per-dataset tools are no longer listed, but a connector set up with them still gets answers. The Manual documents each lookup in detail.

Trust and Limits

• Published values are official statistics or filed figures exactly as the source released them; year-over-year and other rates, ratios and aggregates are calculated by this platform, and every lookup result carries that label plus the formula used.
• Most lookups accept an as of date and answer as the data stood then, from the stored vintage history — so a revised number can be compared with the one that was on the screen at the time.
• Every result includes a permanent citation URL — the page on this site showing the same view — so any figure the assistant quotes can be checked at the source.
• Missing values are returned as missing, never as zero. Filing-derived datasets return their own coverage with every aggregate: they accumulate filing by filing and are not the whole market.
• The assistant's prose is the product of the AI client you connect, not of this site. Check its statements against the cited pages before quoting them; the Methodology page explains every label and formula.
• The endpoint is rate-limited to about 120 lookups a minute per connection. For bulk or programmatic access, use the public API directly.