jpcite vs. subsidy search services and AI evidence APIs

A neutral side-by-side of jpcite against the main consumer-discoverable Japanese subsidy / 助成金 search services, across 13 axes. jpcite is not a generic search portal: it is the Japanese-regulation packet bank that AI agents find and buy: source-linked Evidence Packets (from ¥30) with fetched-at metadata, tamper-evident signatures, and known gaps that an AI fetches before drafting an answer. Built from each provider's published information only; anything we could not verify is left as "Not publicly disclosed" rather than guessed. Reviewed monthly.

Services compared: jpcite (this site) / subsidy search sites / public support portals / article-style databases / J-Net21 / ミラサポplus / Japanese-corporate MCP servers (OSS MCP servers that wrap gBizINFO, EDINET, and e-Stat).

Feature comparison

Axis jpcite Subsidy search site Support portal Article database J-Net21 ミラサポplus Japanese-corporate MCP (gBizINFO/EDINET/e-Stat)
1. Program count (public-sector) 11,601 searchable (tier S/A/B/C) + 14,472 total (incl. rows excluded from public search) Not publicly disclosed (sector-focused) Not publicly disclosed Not publicly disclosed (article-led) Not publicly disclosed (active programs surfaced as feed) Not publicly disclosed (major + related programs) Not publicly disclosed (relays gBizINFO / EDINET / e-Stat raw data; no program catalog count)
2. Primary-source URL on each row (direct gov links) Yes — most public rows carry source_url + fetched_at; gaps are flagged rather than backfilled Not publicly disclosed (needs verification) Not publicly disclosed (needs verification) Not publicly disclosed (needs verification) Yes (links out to ministry / prefecture pages) Yes (.go.jp official portal) Relays government APIs (gBizINFO / EDINET / e-Stat) so the source is the government, but no per-row fetched_at timestamp
3. Public API Yes — REST /v1/* with OpenAPI 3.1 None publicly listed None publicly listed None publicly listed None publicly listed None publicly listed Depends on the underlying government APIs (no proprietary REST spec publicly listed)
4. MCP (Model Context Protocol) server Yes — 261 tools, protocol 2025-06-18, distributed via stdio + Smithery + DXT None publicly listed None publicly listed None publicly listed None publicly listed None publicly listed Yes — OSS MCP server listed on awesome-mcp; callable directly from AI agents
5. Designed for LLM-agent consumption Yes (primary use case) — callable from Claude Desktop / Cursor / Cline via MCP and ChatGPT Custom GPTs via OpenAPI Actions; responses pre-normalized for AI Designed for human web browsing Designed for human web browsing Designed for human web browsing Designed for human web browsing Designed for human web browsing Callable from AI agents via MCP (response normalization / source fencing vary by implementation)
6. Pricing model Per-result pricing: raw evidence from ¥3, Evidence Packets from ¥30 (up to ¥240 deep DD). Weekly free passport. No tier SKUs, no seats, no minimums Free Not publicly disclosed (signup required) Not publicly disclosed (article reading is free) Free (public-sector service) Free (public-sector service) Free (OSS; subject to the rate limits / terms of the upstream government APIs)
7. Free tier No signup; 3 req/day per IP (resets at JST next-day 00:00) Fully free Not publicly disclosed Not publicly disclosed Fully free Fully free Fully free (OSS)
8. Full-text search (Japanese 全文検索 / 分割アルゴリズム) Yes — Japanese full-text search + semantic search In-site search (implementation details not disclosed) In-site search (implementation details not disclosed) In-site search (implementation details not disclosed) In-site search (implementation details not disclosed) In-site search (implementation details not disclosed) Depends on the government APIs' search parameters (no proprietary Japanese full-text / semantic search)
9. Per-row source license disclosure Yes — returned rows carry a license field (e-Gov=CC-BY-4.0, NTA invoice=PDL v1.0, gov_standard, proprietary, etc.) Not publicly disclosed Not publicly disclosed Not publicly disclosed Governed by site Terms (no per-row license metadata) Governed by site Terms (no per-row license metadata) No per-row license field (users check the upstream government data license individually)
10. Statute-text navigation (e-Gov links) Yes — e-Gov law metadata and article references; coverage varies by record; per-article cross-link via program_law_refs Out of scope (subsidy search only) Out of scope (subsidy search only) Out of scope Out of scope (subsidy listings only) Out of scope (subsidy listings only) Out of scope (corporate / financial / statistics data; no statute-text linkage)
11. Adoption / case-study data Yes — 2,286 case_studies (filterable by industry / amount / round) Not publicly disclosed Not publicly disclosed Some examples surfaced editorially Has a 事例 (case studies) section (no API) Has 事例ナビ (Web UI only) Out of scope (no structured adoption-case data)
12. Enforcement-action data Yes — 1,185 enforcement_cases (cross-searchable by ministry / law / sanction type) Out of scope Out of scope Out of scope Out of scope Out of scope Out of scope (no structured enforcement-action data)
13. Corporate lookup (法人番号 / gBiz) Yes — 166,969 corporate records from gBizINFO + 13,801 qualified-invoice registrants (PDL v1.0 delta; full monthly bulk pending) Out of scope Out of scope Out of scope Out of scope (NTA / gBiz are separate services) Out of scope (NTA / gBiz are separate services) Relays gBizINFO, so corporate lookup is possible (no fetched_at / license field attached)

Legend — green: confirmed from public sources / amber: partially present / grey ("None publicly listed"): nothing on the official site / grey italic ("Not publicly disclosed"): could not verify. "None publicly listed" / "Not publicly disclosed" does not assert the feature is absent — providers may add or undocument features at any time. Spot an error? Email [email protected] with a source URL and we'll review it and update if needed.

Pricing 7-band SOT comparison (2026-05-19 operator-approved)

jpcite is an AI-agent-first route that supplies the ingredient (raw material) across 7 tiers. This is a fact-based per-unit comparison against single-shot LLM runs, competitor MCP servers, and pure-LLM web search. jpcite prices mean "ingredient only"; agent finishing (expansion / summarization) runs on the calling-side LLM.

jpcite product line (ingredient granularity) jpcite SOT price (excl. tax) LLM alone (Opus 4.7 single-shot) Pure-LLM web search anchor Competitor MCP (japan-corporate-mcp etc.)
Raw evidence (citation lookup baseline) ¥3 / unit Not possible standalone (the fetch step is embedded inside the agent loop) Not publicly disclosed (no metered API on web search results) Free (public MCP, but no fetched_at / license metadata)
Citation (引用候補) 無料-¥10 Same as above Same as above Free (source URL present, but no license / freshness disclosure)
Verified receipt (T 番号 / 適格事業者検証付) ¥10-30 / receipt Not possible standalone (requires fetching + structuring the NTA public site) Not publicly disclosed Out of scope (NTA PDL v1.0 ingestion confirmed only on our side)
LLM Expansion Kit (素材 packet) ¥30 / packet Not possible standalone (jpcite bundles the prompt kit) Not publicly disclosed Out of scope
Standard Kit (汎用 cohort packet) ¥60 / kit Similar output possible standalone (but missing source URL / fetched_at / license; ~¥120 anchor in agent-side token cost) Retrieving similar information is possible (source accuracy / freshness depend on web search results) Out of scope
Pro DD Kit (深掘り due diligence) ¥120 / kit ¥120 anchor (Opus 4.7 single-shot agent loop; ingredient fetch + finishing included; lacks source fencing) ~¥500 anchor (repeated web search + LLM finishing included; low source accuracy) Out of scope
Deep DD / M&A Bundle (最上位 bundle) ¥240 / bundle Runnable standalone, but no reproducibility / audit trail Same as above Out of scope
Freshness watch (鮮度監視 subscription) ¥50 / 対象 / 月 Not possible standalone (re-run cost = full token every time) Not publicly disclosed Out of scope (no diff-detection mechanism)
Weekly passport (商品ライン毎 free tier) 無料 (first Max Kit 1 + 1 Standard Kit/week, stable-hash re-reads free, counted from Monday 00:00 JST) No equivalent mechanism (full token every time) No equivalent mechanism Being a free service, a passport mechanism is unnecessary (source-accuracy trade-off)

ingredient + finishing boundary: jpcite performs no LLM inference and supplies only the ingredient (raw material). The Opus 4.7 single-shot ¥120 is a finishing-included anchor (the agent loop covers ingredient fetch + structuring + prose). jpcite Pro DD ¥120 is pre-finishing ingredient; adding the calling-side LLM's finishing tokens (~¥10-30 anchor) brings the total to ~¥130-150. For repeated runs and stable-hash re-reads, the weekly passport sharply lowers the per-unit cost. See Pricing.

Which one should you pick?

Honest recommendations by use case. jpcite is not always the right answer.

SMB / Business owner

Start with the free public portals

If you only need to find one or a few subsidies for your own company, the SMRJ-operated portals — ミラサポplus and J-Net21 — are free, official, and a perfectly fine starting point.

If you want an LLM to read the public-call PDF and check eligibility against your business, filter case studies by industry, or pre-screen before handing things to a tax accountant, then jpcite's API (¥3/billable unit, 3 req/day free without signup) is worth a try.

Tax / legal / consulting professionals

Across many clients, primary-source linkage is mandatory

If you serve 10+ client companies, citing stale amounts or already-ended programs from secondary aggregators creates real liability (景表法 / 善管注意義務 risk). You need primary-source URLs and fetch timestamps on returned public rows.

jpcite ships source_url + fetched_at on major public rows, direct e-Gov statute links, and 1,185 enforcement cases (vendor due diligence) — callable from Claude Desktop / Cursor / Cline via MCP and ChatGPT via OpenAPI Actions.

Developers / LLM-agent builders

If you need API/MCP, your shortlist is short

Most subsidy search sites and public support portals do not publish open APIs or MCP servers as of 2026-04-26. Calling them from an AI agent means HTML scraping plus per-site Terms review.

jpcite ships REST + MCP (protocol 2025-06-18) + OpenAPI 3.1 + 261 tools at standard configuration. Evaluate with 3 free req/day, then pay per result (raw evidence from ¥3, Evidence Packets from ¥30) — no SDK, no seats.

Update policy

Related pages