Japan Diet Records Search MCP

Search Japan's National Diet (parliament) proceedings — speeches and meeting records — via the National Diet Library's official Kokkai (Diet Records) Search API, exposed as MCP tools for LLM agents (Claude, Cursor, etc.) plus a normal batch-run mode. This Actor is an independent tool and is not affiliated with or endorsed by the National Diet Library or the Diet.

国立国会図書館「国会会議録検索システム 検索用API」を使い、国会の発言・会議記録をLLMエージェント（Claude、Cursor等）から検索できるMCPツール群として提供するActorです。通常run（バッチ）呼び出しにも対応します。本Actorは非公式のツールであり、国立国会図書館や国会とは無関係です。

What does this Actor do?

• Runs as an Apify Standby MCP server exposing 4 tools: search_speeches, search_meetings, get_speech, summarize_topic
• Also runs as a normal (batch) Actor: give it JSON input with a mode field and it pushes results to a dataset, like any other Actor
• Caches search results and individual speech records for 24 hours to avoid repeat requests to the source API
• Returns excerpts (300 characters) with source URLs by default; full text is available per-speech (up to 5,000 characters) via get_speech, respecting the speaker's copyright in the underlying transcript

Why use it?

Existing community MCP servers for this API exist but are unmaintained (the one listed Actor on Apify Store has a 0% success rate). This Actor wraps the same official, no-API-key-required NDL search API with request pacing, response-size safety limits, and result caching, and ships both as an MCP server and a regular Actor.

How to use it — as an MCP server

1. Deploy this Actor (or use the published one) and note its Standby URL, e.g. https://<username>--kokkai-search-mcp.apify.actor/mcp.
2. Add it to your MCP client (Claude Desktop, Claude Code, Cursor, etc.) as a Streamable HTTP MCP server, passing your Apify API token as a Bearer token in the Authorization header.
3. Ask your LLM to search Diet speeches — it will call search_speeches / search_meetings / get_speech / summarize_topic as needed.

How to use it — as a normal run

Give it input like:

{ "mode": "search_speeches", "query": "消費税", "limit": 10 }

and it pushes the results to the default dataset, same as any batch Actor. Leave mode empty (or omit it) to start the MCP Standby server instead.

Input

Field	Type	Description
mode	string	Leave empty to run as an MCP Standby server. Otherwise: search_speeches, search_meetings, get_speech, or summarize_topic.
query	string	Full-text search term. One of query / speaker / from / until is required by the underlying API.
speaker	string	Speaker name (search_speeches / summarize_topic only; ignored by search_meetings, since meetings aren't tied to a single speaker).
house	string	衆議院 / 参議院 / 両院 / 両院協議会.
from / until	string (YYYY-MM-DD)	Meeting date range.
limit	integer	Max records (API hard limit: 100).
speechId	string	Required for get_speech — must come from a prior search_speeches/summarize_topic result (the API has no direct speechId lookup).
maxChars	integer	Max characters for get_speech full text (up to 5,000).

Output

{
  "speechId": "122105261X01720260727_073",
  "meeting": "予算委員会",
  "house": "衆議院",
  "date": "2026-07-27",
  "speaker": "後藤祐一",
  "snippet": "○後藤（祐）委員　ブロックチェーンという技術で、暗号資産は…",
  "url": "https://kokkai.ndl.go.jp/txt/122105261X01720260727/73"
}

You can download the dataset in various formats such as JSON, HTML, CSV, or Excel from the Output tab.

Pricing

Pay-per-event:

Event	Price	When
tool-call	$0.005	Each MCP tool call (cached results are still charged)
actor-start	$0.02	Once per normal (non-MCP) run
speech-record	$0.001	Each record pushed to the dataset in a normal run

This Actor runs in Standby mode so the MCP server is ready without a cold start. As the caller, you are only ever charged the pay-per-event prices above — for a tool call or a dataset item pushed — never for the Actor simply sitting idle waiting for requests.

Tips

• Results are cached for 24 hours per unique query, so repeated questions about the same topic within a session don't re-hit the source API or incur extra tool-call latency (though each MCP call is still charged — see above).
• get_speech only works for a speechId returned by a prior search_speeches or summarize_topic call in the same session, since the underlying API has no way to look up a speech by ID directly.
• This Actor does not call an LLM itself (no external API key needed) — summarize_topic just returns a larger batch of excerpts for your own LLM to summarize.

FAQ, disclaimers, and support

• Speech text is subject to the speaker's own copyright; this Actor returns short excerpts by default and caps full-text retrieval at 5,000 characters per speech for information-analysis purposes.
• This Actor only uses the National Diet Library's official public search API (no login, no API key, no scraping).
• This Actor only searches public, official proceedings of sitting members of the Diet (public figures' public statements). It is not intended for — and should not be used for — collecting personal information about private individuals; it stores no personal data of its own beyond a 24-hour cache of NDL search results and speech records used to make get_speech work. Expired cache entries are removed on a best-effort basis (on read, and via a bounded background sweep) rather than guaranteed to disappear the instant they expire.
• Found a bug or have a feature request? Open an issue on the Actor's Issues tab.