dsh-cn-disclosure

A-share disclosures — earnings previews, buybacks, dividends, lock-up expiries, litigation, M&A deals — carry dense quantitative detail, yet every company formats its filings differently. Pulling out the right fields (what happened, when, how much, how many shares) from free text has long been a manual bottleneck in event-study pipelines. Most existing A-share MCP servers focus on fetching and summarizing announcements, while local-first structured extraction remains almost entirely unoccupied. dsh-cn-disclosure fills exactly that gap: it does not scrape, does not call external APIs, and has one focused job — turning a passage of disclosure text into event-research-ready structured fields.

Built on pure Node ESM with zero npm dependencies, the plugin runs fully offline, keeping documents and any PII on the machine at all times. It ships with 17 disclosure event types (earnings preview, dividends, buybacks, insider trading, lock-up expiry, M&A, private placements, guarantees, litigation, contract awards, equity incentives, executive changes, and more) and 9 field-extraction patterns (dates, amounts in yi/wan, percentages, share counts, EPS, per-share dividend, and others). Instead of assuming a fixed template, regex scanning of free text tolerates the wide formatting variety across Chinese listed companies. Five tools cover the full pipeline — classification, full structured extraction, event listing, local file ingestion (txt, md, html with tag stripping), and rule-set introspection — all returning standard UTF-8 JSON for downstream consumption. The entire rule set lives in a single JSON file; adding new event types or field patterns requires no code changes, just a server restart.

If you build A-share event studies, factor models, or earnings-tracking pipelines and need to quickly structure a snippet of announcement or annual-report text into quantifiable fields — or if data compliance mandates that disclosure documents never leave your machine — this plugin is designed for you. It does not replace announcement-fetching tools; it acts as a local structured-extraction layer that slots cleanly into your existing research and data workflows.