Retrieve Japanese regulations, tax, and law reliably — through your AI.

Connect your business AI (ChatGPT / Claude.ai / Perplexity) or developer AI (Cursor / Claude Code / Bedrock) to jpcite. Pay ¥30 per Evidence Packet — primary-source URLs, fetched_at timestamps, and tamper-evident signatures included — and fetch instantly. Your AI returns a cited answer in about 30 seconds.

Try free (3 requests/day per IP, no signup) Add to Claude Desktop (.mcpb)

Free anonymous evaluation, 3 requests/day per IP (JST UTC+9 next-day 00:00 reset). Evidence Packets from ¥30 each; raw evidence from ¥3. Primary-source URL + fetched_at on most public rows.

Data & quality

503,930 entities + 6.12M facts

9 datasets + structured entity facts, one query surface

Programs, adoption case studies, loans, enforcement, laws, court decisions, tax rulesets, bids, invoice registrants — all under a unified schema. The entity-fact layer carries entities and facts across 12 record kinds. Japanese search lets you query terms as written, no romanization required.

  • programs · 14,472 (11,601 searchable = Tier S 114 + A 1,340 + B 4,186 + C 5,961; 2,871 excluded from public search)
  • case_studies · 2,286
  • loan_programs · 108 (collateral / personal-guarantor / third-party-guarantor decomposed)
  • enforcement_cases · 1,185
  • laws · metadata, searchable references, and e-Gov provenance for compliance lookup
  • tax_rulesets · 50
  • court_decisions · 2,065
  • bids · 362
  • invoice_registrants · 13,801 (NTA PDL v1.0 delta; full monthly bulk pending)
  • structured entity facts · 503,930 entities / 6.12M facts / 378,342 relations / 335,605 aliases

Primary-source trace

source_url + fetched_at on most public rows

Sourced directly from METI, MAFF, the SME Agency, Japan Finance Corporation, prefectures, and municipalities. Aggregator sites are excluded by policy. Designed to track amendment signals and refetch status continuously.

  • second-tier aggregators excluded
  • Nightly source-URL liveness scan
  • Lineage columns tracked for searchable rows

181 exclusion rules

Tier S/A/B/C/X quality labels

Co-use conflicts, prerequisites, and mutual exclusivity are encoded as structured rules. Post a list of program_ids, get back the conflicts. No "plausible-sounding" LLM inference — actual rules.

  • 125 exclude + 17 prerequisite
  • 15 absolute + 24 other
  • Tier S/A/B/C = data-completeness ranks; non-public rows are excluded from search

v0.4.0 output surface

jpcite is a Japanese-regulation packet bank that acts as a context-compression layer before an AI reads Japanese institutional evidence. REST, MCP, OpenAPI Actions, widgets, webhooks, and datasets are entry surfaces for saved outputs a user can inspect, hand off, or automate. See the connect guide for AI agent developers (Claude / Cursor / ChatGPT / Codex).

Evidence Packet

For answer generation and audit trails

Primary-source URLs, fetched_at, known gaps, professional-scope fences, and citation candidates in JSON or Markdown.

Company Folder

For one-company checks

Corporate-number or invoice-number lookup with identity match, invoice status, enforcement records, adoption history, and next questions.

Company Public Fact Brief

For M&A, lending, and counterparty review

Public records, enforcement history, subsidy adoption, unverified scope, and analyst notes separated for review.

Monthly Client Review

For recurring advisory work

Regenerates program candidates, tax changes, deadlines, and client-question drafts by client, industry, and region.

Application Evidence Pack

For pre-application meetings

Turns region, industry, investment amount, and use of funds into candidate programs, exclusion rules, stacking cautions, required docs, and interview questions.

Funding Compatibility

For checking combinations before advice

Turns program IDs or candidate programs into mutual exclusions, missing prerequisites, unknowns, and source material to verify.

Invoice Counterparty Check

For accounting and tax review prep

Turns invoice numbers, corporate numbers, or vendor CSV rows into registration status, identity matches, history, and rows needing review.

Funding Traceback / Source Receipt

For proving what was checked

Records fetched time, content hash, known gaps, and recheck targets for a source URL, program ID, law ID, or article ID.

Law / Case Citation Candidates

For citation-first AI answers

Returns candidate law articles, related cases, and professional-review notes from law names, articles, programs, or issue keywords.

Amendment/Saved Search Alerts

For watchlists

Delivers law amendments, program changes, saved-search diffs, notification logs, and webhook events.

Cost/Payment Control

For automated agents

Usage, caps, duplicate-charge prevention, payment verification, and retry status via API keys, headers, wallets, or x402.

Agent Handoff

For Claude, Cursor, ChatGPT, Codex

Shares the same evidence across REST, MCP, and OpenAPI Actions so multiple agents work from one source-backed surface.

Products Preview an Evidence Packet in the live playground

Why this can reduce repeated document-token cost

Send the AI a compact Evidence Packet instead of raw long-context source text.

jpcite does not replace the model's answer generation. It replaces repeated retrieval stuffing: source URLs, fetched_at timestamps, known gaps, and citation candidates are returned as a smaller packet that Claude, ChatGPT, Cursor, Codex, or an internal workflow can read before answering.

Raw long-document input

Cost is driven by source-document tokens sent to the external model, plus output tokens.

jpcite

Cost is ¥3 per billable unit plus the smaller token cost of reading the returned Evidence Packet.

Do not claim jpcite is always cheaper. It is strongest for repeated Japanese public-record checks, batch DD, saved searches, and agent workflows where long official source text would otherwise be re-sent to the LLM.

What each path returns

Start from the user's input: a natural-language query, corporate number, saved topic, or website visitor form. jpcite returns source-linked candidates, rules, alerts, or handoff context.

API + MCP

AI developers / enterprise evidence workflows

Call 261 MCP tools directly from an agent. Prefetch Evidence Packets with primary-source URLs before answer generation.

  • ¥3/billable unit tax exclusive (¥3.30 tax inclusive)
  • Anonymous 3 requests/day per IP free
  • Claude Desktop / Cursor via MCP; ChatGPT via Actions

Read the docs →

Email notifications

SMB owners / professional advisors

Email notifications keep saved searches, deadline reminders, and consultation-pack updates visible after a prescreen is created through the web, API, or widget.

  • Input: saved search or consultation pack
  • Output: deadline reminder, handoff update, review prompt
  • Email is a reminder channel, not a paid questioning surface

Set up notifications →

Law-amendment alerts

Tax / labor / compliance professionals

Email notifications on amendments to invoice law, electronic-bookkeeping law, labor standards, and other tracked domains. Catch changes before deadlines bite.

  • Outputs source-linked amendment evidence before drafting client updates
  • Monthly digest and real-time alerts are delivery channels
  • Billable API work follows the published ¥3/billable unit model

Configure alerts →

Embeddable widget

Tax-firm sites / chambers of commerce / advisory offices

Add subsidy search to your own site with a single <script> tag. Visitors narrow down candidates without leaving your page.

  • Anonymous 3 requests/day per IP free
  • Paid usage follows the published packet pricing, raw evidence from ¥3
  • Widget responses preserve source URLs and known gaps

Embed the widget →

Review handoff

Certified support orgs / tax / labor / SMB consultants

Evidence-to-expert handoff for cases that need licensed or human review. jpcite prepares the source packet and open questions; the professional judgment stays outside the API.

  • Includes known gaps and source receipts
  • Does not decide tax, legal, labor, audit, or filing outcomes
  • Designed for review handoff, not automated advice

Register as an advisor →

Not sure which fits?

Comparison table + decision tree

Individual / corporate / developer / advisor — a short decision tree maps your input to the result you need. Typical use cases and setup time line up side by side.

  • Input-to-output comparison table
  • Decision tree per audience
  • Recommended paths by workflow

See the comparison →

Five ways to use it — pitch by audience

Same backend, different framing for each role.

Tax advisors (zeirishi)

Mid-size firm, ~80 SMB clients

When a client asks "which subsidies apply to us this year?", point Claude at the API. Buy raw evidence from ¥3 or Evidence Packets from ¥30 — around ¥3,000/month for a typical firm. Email digest of monthly amendments included.

  • API + MCP (¥3/billable unit)
  • Law-amendment alerts as delivery channels
  • Monthly digest is included

Tax advisor details →

Administrative scriveners (gyoseishoshi)

Construction-focused, permit-heavy caseload

Pull "applicable subsidies + loans + permits" for a case in one call. Filter by region × industry × deadline. Try the first 3/day free.

  • API + MCP — one call, cross-source
  • Anonymous 3 requests/day per IP free
  • Region × industry × deadline search

Admin scrivener details →

SMB owners

Small business — owner, spouse, one bookkeeper

Run a web prescreen for "subsidies for my industry", keep the source URLs and open questions, then use email reminders for saved searches and handoff updates.

  • Input: prefecture, industry, planned investment
  • Output: candidate programs, source URLs, handoff notes
  • Email notifications for saved-search reminders

SMB details → · Email notifications

VC / M&A advisors

Investors / acquirers running a DD pipeline

From a corporate number, fetch 5 years of enforcement history, 10 years of adoption history, and invoice-registrant status in one query. Drop straight into your DD pipeline. Raw evidence from ¥3, Evidence Packets from ¥30, up to ¥240 for a Deep DD packet — batch-friendly within API rate limits.

  • One query keyed on corporate number
  • Enforcement 1,185 / adoptions 2,286 / invoice 13,801
  • Raw evidence from ¥3 up to ¥240 packets; budget caps and rate limits apply

VC / DD details →

AI agent developers

Claude Desktop / Cursor / your own SaaS integration

A Japanese subsidy / tax / loan MCP server. 261 tools. ¥3/billable unit, 3/day free. Works in Claude Desktop and Cursor out of the box.

  • 261 MCP tools, protocol 2025-06-18
  • stdio transport, no SDK
  • 3 requests/day evaluation, then raw evidence from ¥3 or packets from ¥30

Developer details → · MCP tool list

Not on this list?

If none of the five fit

Labor consultants, SMB consultants, accounting SaaS, chambers of commerce, municipal officers — the decision tree finds them too. Compare typical use cases, combo recommendations, and setup time side by side.

  • 5-question decision tree
  • Recommended paths by workflow
  • Typical use cases lined up

Audience comparison →

Try in 5 seconds (no signup)

Anonymous IP gets 3 requests/day free. Paste and run.

$ curl 'https://api.jpcite.com/v1/programs/search?q=IT%E5%B0%8E%E5%85%A5&limit=3'
{
 "results": [...],
 "total": 17
}
Query is URL-encoded (q=IT導入).

Using an MCP client? → Agent setup guides · Claude Code · Cursor · Codex

This is a source-linked evidence/output API. We do not provide legal advice, tax advice, or filing services. Activities reserved under 弁護士法 § 72 / 税理士法 § 52 / 行政書士法 § 1の2 are out of scope. See error_handling · pricing · Terms.

Subsidy Prescreen — Top 5 in 3 questions

Pick prefecture / business form / planned investment. We'll show the top 5 of 11,601 active programs ranked by fit. Free, no signup, anonymous (3 requests/day/IP). One call to /v1/programs/prescreen.

Picking "Nationwide" drops the regional filter and searches every program. Picking a specific prefecture matches its prefecture-only programs plus all nationwide programs.
"Unspecified" matches only programs accepting both forms. Picking "Corporation" or "Sole proprietor" returns form-specific programs plus shared ones.
Planned investment for equipment, IT systems etc., in 万円 units (optional). Example: 800 = 8,000,000 yen. Programs with ceilings near your amount are ranked higher.

Matching is rule-based against the program database, not LLM inference (prefecture + target_types + amount_fit). Source URLs are primary only. Submissions are not stored — only an IP hash is logged for rate limiting (resets at JST next-day 00:00).

Why use jpcite before asking an AI agent?

AI agents can write quickly, but Japanese public-program work still needs current source links, fetched-at metadata, and known gaps before the final answer is drafted.

Stale training data

Calendar mismatch

Japanese subsidies move on calendar deadlines ("application closed yesterday"). An AI workflow may need current source records and fetched_at metadata before it drafts a client-facing answer. jpcite ships fetched_at on searchable rows where available.

Source fabrication & aggregator drift

Misidentification creates review work

Second-tier aggregator pages can list expired programs as active, mis-state ceiling amounts, or add eligibility requirements that are not in the primary notice. Using that material without checking primary sources can create client-facing errors and extra review work.

jpcite bans aggregator URLs from source_url as a hard policy. All sources are primary: ministries (METI / MAFF / SME Agency), prefectures, government finance corps (JFC / SMRJ / NEDO), and chambers of commerce. Most public rows carry a primary URL; the remaining 12 (small-municipality programs) are flagged "primary URL missing" rather than backfilled with aggregators.

Most public rows carry source_url + fetched_at, so you can show your client or reviewer exactly why this program matched. That source trail is harder to preserve if aggregators are in the chain.

No compatibility logic

Combination rules can't be inferred

LLMs can't structurally answer "can I combine subsidy A with subsidy B?" or "does this conflict with my prior award?". jpcite ships 181 exclusion / prerequisite rules for machine-decidable compatibility.

In short: AI workflow + jpcite = faster source review with primary-source links. jpcite isn't an LLM replacement — it's the evidence layer that makes AI output easier to cite and review.

Why jpcite

No aggregator sources

Primary-source review support

Second-tier aggregator pages can keep expired programs listed as active, mis-state ceiling amounts, or add eligibility requirements that are not in the primary notice. Relying on that material without primary-source review can create client-facing errors.

jpcite's policy is to never register an aggregator URL in source_url. Sources are restricted to ministries (METI / MAFF / SME Agency), prefectures, government finance corps (JFC / SMRJ / NEDO), and chambers of commerce. Most public rows carry a primary URL; the remaining 12 (small-municipality programs without a dedicated CMS page) are flagged "primary URL missing" rather than backfilled.

Because most public rows carry source_url + fetched_at, you can show your client or reviewer exactly why a program matched. That source trail is harder to preserve if aggregators are in the chain.

Primary source, end to end

Fetch dates and liveness are explicit

Most returned public rows carry source_url + fetched_at. The fetch date is shown verbatim — never relabeled as "last updated" (which would falsely imply we re-verified currency). Nightly liveness scans flag dead URLs.

Exclusion logic is in the data, not the prompt

Catch ineligible combinations up front

181 exclusion / prerequisite / absolute rules, structured. The "agent merrily files two incompatible subsidies, both get rejected months later" failure mode is detected the moment you pass program_ids in.

Cross-dataset glue

One call: program → law → court decision → enforcement

Cross-dataset lookups cover laws (e-Gov, CC-BY; metadata, searchable references, and provenance), tax rulesets (invoice law / electronic-bookkeeping law), NTA invoice registrants, court decisions, bids, and structured entity facts. The MCP surface includes trace_program_to_law / combined_compliance_check / search_tax_incentives / rule_engine_check for cross-domain resolution.

Who is it for

Developers / AI engineers

Recommended: API + MCP

For agents, review workflows, or your own SaaS. OpenAPI 3.1, stdio MCP, and Evidence Packets for citation-first workflows; token and cost impact is workload-dependent.

SMB owners

Recommended: email notifications + review handoff

Want to know what your business is eligible for, and to be routed to someone who can file it. Run a web or API prescreen, keep source URLs and open questions, then receive email reminders and handoff updates.

Professional advisors (tax / labor / SMB consultants)

Recommended: alerts + embeddable widget + review handoff

Don't miss amendments, prepare source-linked Evidence Packets, and hand off open questions to licensed review when needed. Billing follows the published ¥3/billable unit model.

Corporate (compliance / accounting)

Recommended: alerts + API for private knowledge workflows

Stay current on invoice law, electronic-bookkeeping law, and Premiums & Misrepresentation Act amendments. Alerts surface the change; the API serves the detail to your own tools.

Pricing summary

Interface Free tier Paid Tax inclusive
API + MCP 3 requests/day per IP ¥3 / unit (tax excl.) ¥3.30 / unit
Email notifications Reminder channel Not a paid questioning surface Saved-search and deadline reminders
Law-amendment alerts Monthly digest ¥3 / notification (tax excl.) ¥3.30 / notification
Embeddable widget 3 requests/day per IP ¥3 / search (tax excl.) ¥3.30 / search
Advisor affiliate Free for end users ¥3,000 / signed engagement (paid by professional) ¥3,300 / engagement

Full pricing details → · Auto-issued qualified invoice via Stripe Tax; cancel anytime. Billing is JPY; 10% JCT added at invoice; your card network handles FX.