HKEx Filings MCP Server

HKEx Filing Scraper

An open-source Python tool that scrapes 25+ years of Hong Kong Stock Exchange (HKEx) regulatory filings and ingests them into any combination of nine databases — with full-text and table extraction, chunk-level coverage, optional graph linking, and a read-only MCP server so AI agents can query the corpus or the live site.

<!-- mcp-name: io.github.simonplmak-cloud/hkex-filings -->

It speaks the undocumented HKEx JSON API directly, which is faster and more resilient than driving a browser.

Vendors & integrations

Databases — nine first-class destinations, in documented popularity order (see the support matrix):

• PostgreSQL — production-grade open-source relational
• MySQL / MariaDB — GPL relational servers, one driver
• SQLite — zero-server file database, no install needed
• MongoDB — document database
• Neo4j — property-graph database
• ClickHouse — columnar analytics engine
• DuckDB — in-process analytical engine
• SurrealDB — multi-model graph + document database

AI clients — any MCP-capable agent; ready-made configuration for Claude, ChatGPT, Cursor, VS Code/Copilot, Gemini CLI, opencode, Manus, and Perplexity.

Available on — PyPI · Glama · MCP Registry · hosted gateway.

Two ways to use it

	Hosted MCP gateway	Local pipeline
What	A public endpoint you point an AI agent at	The hkex-scraper CLI
Setup	None — paste a URL	pip install + one environment variable
Data	Live from HKEx, nothing stored	Stored in your database(s)
Docs	Live MCP gateway · AI agent support	Getting started

Use the hosted MCP gateway

POST, Streamable HTTP, no API key:

https://hkex-listco-updates.ascent-partners.com/api/mcp

Four read-only tools: get_server_info, search_filings (a window of at most 31 days, with optional stock-code, title, document-type, category, and stock-name filters), list_filing_facets (browse what a window contains), and get_filing (downloads one document and extracts its text and tables).

Point a client at it — for example opencode:

{
  "$schema": "https://opencode.ai/config.json",
  "mcp": {
    "hkex-live": {
      "type": "remote",
      "url": "https://hkex-listco-updates.ascent-partners.com/api/mcp"
    }
  }
}

Then ask:

Use hkex-live to list the filings published between 2026-09-01 and 2026-09-18,
then summarise the interim report.

Ready-made configuration for Claude, ChatGPT, Cursor, VS Code/Copilot, Gemini CLI, opencode, Manus, and Perplexity is in AI agent support — and for a stored corpus, the stdio MCP server exposes a wider tool catalog and is published on Glama. The gateway is listed in the official MCP Registry as io.github.simonplmak-cloud/hkex-filings.

Featured on Glama — the read-only stdio MCP server is also published on Glama, where Glama scans the built server and scores tool-definition quality (currently 4.7/5).

Quick start (local)

pip install hkex-filing-scraper        # core; SQLite needs no server
pip install "hkex-filing-scraper[all]" # Excel + dotenv + every driver + the MCP server
cp .env.example .env                   # then set DATABASE_TARGET (below)
hkex-scraper --metadata-only --limit 100

Optional extras: excel, postgres, mysql, duckdb, mongodb, clickhouse, neo4j, mcp, pdf, all, dev.

DATABASE_TARGET is an ordered, comma-separated list of sink ids; the order decides which sink serves reads. To start with no server:

DATABASE_TARGET=sqlite
SQLITE_PATH=hkex.db

hkex-scraper runs the full pipeline (metadata + documents + graph); hkex-scraper --full-history covers everything since April 1999. The schema is created automatically. Full install options and per-sink settings are in Getting started.

Database support

Every sink is a first-class destination; rows are in documented popularity order. The full matrix — licenses, capability differences, per-engine notes — is in Database sinks.

Sink	Model	License	Extra	Idempotent upsert
postgres	relational	PostgreSQL License	postgres	ON CONFLICT DO UPDATE
mysql / mariadb	relational	GPLv2	mysql	ON DUPLICATE KEY UPDATE
sqlite	relational	Public domain	—	ON CONFLICT DO UPDATE
mongodb	document	SSPL¹	mongodb	update_one(upsert=True)
neo4j	graph	GPLv3 (Community)	neo4j	MERGE
clickhouse	columnar	Apache-2.0	clickhouse	ReplacingMergeTree + read-merge
duckdb	relational	MIT	duckdb	ON CONFLICT DO UPDATE
surrealdb	graph + document	BSL 1.1¹	—	UPSERT / RELATE

¹ Source-available, not OSI-approved — labeled exceptions per ADR 0003.

Valid sink ids, in documented order: postgres, mysql, sqlite, mongodb, mariadb, neo4j, clickhouse, duckdb, surrealdb. Set one variable and the same run feeds every sink:

# Order sets read precedence.
DATABASE_TARGET=postgres,sqlite
POSTGRES_DSN=postgresql://user:password@localhost:5432/hkex
SQLITE_PATH=hkex.db

How it works

flowchart LR
    A[HKEx JSON API] --> B[Phase 1: metadata]
    B --> C[Canonical record]
    C --> D{DATABASE_TARGET}
    D --> E[(PostgreSQL)]
    D --> F[(MySQL / MariaDB)]
    D --> G[(SQLite)]
    D --> H[(MongoDB)]
    D --> I[(Neo4j)]
    D --> J[(ClickHouse)]
    D --> K[(DuckDB)]
    D --> L[(SurrealDB)]
    B --> M[Graph linking]
    M --> D
    B --> N[Phase 2: download and extract]
    N --> C

• Phase 1 scrapes filing metadata through a JSF session, splitting the range into monthly chunks and deduplicating on a 16-character MD5 filingId.
• Phase 2 downloads each filing's PDF/HTML/Excel document, extracts text and tables to Markdown, and writes the payload.
• Graph linking (optional) writes has_filing and references_filing edges when COMPANY_TABLE is set.
• Failure isolation — a failure on one sink is logged and counted but never blocks another; the run exits non-zero if any configured sink failed.

Deeper detail: Architecture · ADR 0002.

Features

• Fast API scraping — direct HKEx JSON API; no browser or Selenium.
• Full history — every filing from April 1999 to today, with chunk-level coverage checks.
• Document processing — PDF/HTML/Excel text and structured tables, extracted to Markdown.
• Multi-sink — any ordered combination of nine databases, each with native idempotent upserts.
• AI-ready — a hosted live MCP gateway plus a local stdio MCP server.
• Resumable and observable — batching, parallel downloads, stalled-job detection, per-sink counters, and --coverage-report / --parity-report / --verify.
• Optional dependencies — the core is requests + beautifulsoup4; drivers and document extraction are extras with graceful fallbacks.

Documentation

• Getting started · Configuration · CLI
• Database sinks (matrix) — PostgreSQL, MySQL/MariaDB, SQLite, MongoDB, Neo4j, ClickHouse, DuckDB, SurrealDB
• Live MCP gateway · AI agent support · MCP server
• Architecture · Troubleshooting · Testing
• Roadmap · De-risking register · Upgrading
• What's new · Releasing · Legal & Terms of Use · Changelog
• Docs site: https://hkex-listco-updates.ascent-partners.com/ · Try it locally (examples/)

Development

pip install -e ".[dev,all]"
ruff check           # lint (py310, line-length 100)
ruff format --check  # formatting
pytest               # unit tests (no DB or network required)

Tests are pure unit tests; SQLite and DuckDB contract tests run in-process, and integration tests that need a server are skipped unless that sink is configured. See Testing.

Contributing

See CONTRIBUTING.md; report security issues per SECURITY.md. Ideas and questions are welcome in Discussions.

If this saves you time, a star helps others find it.

License

MIT — see LICENSE. That covers this project's code only; optional dependencies carry their own licenses, notably the pdf extra (PyMuPDF / pymupdf4llm), which is AGPL-3.0 and deliberately excluded from .[all]. See docs/legal.md.

Data & Terms of Use: this is a research tool for the undocumented HKEx JSON API, and it is not affiliated with or endorsed by HKEx. Commercial redistribution of HKEx data may require a licensed HKEx feed; see docs/legal.md.