天启至数 Apocdata — A股数据服务 (Apocdata A-Share Data Skill)

English | 简体中文

Let AI read A-share announcements

The A-share announcements and fundamentals database built for AI · parsed · structured · traceable

All-market announcements parsed into structured data: AI summary · category · importance · sentiment, fully traceable to source. Also 45+ endpoints for quotes, financials, capital flows, dragon-tiger lists and macro data. Free tier for instant access; register for higher quotas. Compatible with Cursor, Codex, Hermes, DeepSeek, Kimi, Qwen, Zhipu and any AI agent that supports tool calling.

The SKILL.md in this repository is a capability card that can
be loaded directly into AI agents, letting LLMs query A-share market data
and complete research tasks without any SDK.

Quick Try

Copy-paste and run, no config needed:

curl -s "https://www.apocdata.com/api/blade-dataplatform/open/data/quote?symbol=600519"

{
  "code": 200, "success": true,
  "data": {
    "symbol": "600519", "name": "Kweichow Moutai",
    "close": 1528.00, "pct_chg": -0.52,
    "volume": 2856321, "delayed_minutes": 15
  }
}

Actual response has more fields; above is a simplified example.

Announcement Parsing — from PDF to structured data

Conventional data sources hand you the raw announcement text and leave the reading and parsing to you. Apocdata returns the parsed, structured result, ready for an AI agent to consume.

Every announcement carries:

• AI summary (summary) — the key facts extracted, no need to read the full document
• Category (category) — earnings, dividend, buyback, share increase/decrease, litigation and more, filterable programmatically
• Importance level (importance) — locate the announcements that actually matter
• Sentiment (sentiment) — positive / negative / neutral, to help judge the direction of impact
• Source link (url) — every conclusion traces back to the original announcement

Also returned: title, ann_date, publish_time, keywords, source, and
content (full Markdown body, on includeContent=true).

Covers all-market announcements, including Stock Connect constituents, ST and delisting-risk securities.

Why Apocdata?

Comparison based on publicly available info as of 2026-08.

Core Capabilities

• Announcement parsing — all-market announcements parsed by AI into structured fields
  (title / category / importance / ann_date / summary / sentiment / url), directly
  usable for reasoning and traceable to the source document (mainstream data sources mostly ship
  announcements as raw text)
• Full-dimension coverage — quotes, financials, valuation, capital flows, limit-up review,
  dragon-tiger lists, sectors & concepts, convertible bonds, macro indicators and trading
  calendar: 11 groups, 45+ endpoints, one consistent definition set
• Multiple integration paths — Skill package (one-command install), MCP server (46 tools),
  plain HTTP endpoints (direct curl), covering every mainstream AI ecosystem
• Traceable — every data point traces back to the original announcement, so AI conclusions
  can be checked

Table of Contents

• Announcement Parsing
• Why Apocdata?
• Core Capabilities
• Overview
• Use Cases
• Data Service Platform
• Installation
• Project Structure
• Basic Usage
• Endpoint Overview
• Editions & Capabilities
• Notes & Compliance

Overview

Apocdata is the A-share announcements and data layer built for AI: parsed,
structured, traceable. Announcement parsing is the core — every announcement
becomes structured data an agent can reason over directly — and around it sits
full-dimension coverage of quotes, financials, capital flows, dragon-tiger
lists, quantitative factors and macro indicators, designed for AI agents,
quantitative research, investment content, and institutional applications.

Use Cases

• "What are the key points in Kweichow Moutai's recent earnings flash?" → key
  figures extracted as structured fields, traceable to the original announcement
• "Analyse the valuation of 688017 — how does it compare with its peers?" →
  PE/PB/market cap aggregated with financial metrics and the last 30 sessions
• "Which ST stocks published material announcements in the past 30 days?" →
  ranked by risk level, each with an AI summary and a source link
• "Which sectors has northbound capital been flowing into?" → Stock Connect
  plus sector capital flows, with the top inflow sectors summarised automatically

Data Service Platform

Platform URL: https://www.apocdata.com

The data service platform is the unified entry point for Apocdata. It provides:

• Open API (free tier) — This is what the Skill connects to. A free trial
  quota is callable straight away, ideal for quick validation and lightweight usage.
• OpenAPI documentation & SDKs — Python / TypeScript supported.
• Registered & Professional editions (with API key) — Higher quotas, deeper history,
  lower latency, and richer data fields. See Editions & Capabilities.

The base URL used by this Skill: https://www.apocdata.com/api/blade-dataplatform/open/data

Installation

Quick Install (recommended)

mkdir -p ~/.claude/skills/apocdata
curl -sL https://github.com/ApocData/ApocData-skill/archive/refs/tags/v2.0.5.tar.gz \
  | tar xz -C ~/.claude/skills/apocdata --strip-components=1

Restart Claude Code and the skill will be auto-detected.

MCP Install (Claude Desktop / Cursor / ChatGPT)

{
  "mcpServers": {
    "apocdata": {
      "command": "npx",
      "args": ["-y", "@apocdata-info/mcp-server"]
    }
  }
}

OpenAPI Import (GPT Actions / Dify / Coze / n8n)

https://www.apocdata.com/api/blade-dataplatform/open/data/openapi.json

Import and use — free tier, no registration needed.

Alternative: install script

curl -sL https://raw.githubusercontent.com/ApocData/ApocData-skill/v2.0.5/scripts/install.sh | bash

Project Structure

This Skill uses a multi-file structure for progressive loading — the entry
SKILL.md is a slim router (~180 lines), and detailed endpoint specs are
loaded on demand from the references/ directory:

├── SKILL.md                  # Entry point (slim router, ~180 lines)
├── README.md                 # This file (human-facing docs)
├── CHANGELOG.md              # Version history
├── references/               # On-demand reference docs (loaded per topic)
│   ├── boundaries.md         # Interface boundaries, headers, error codes, cache, freshness
│   ├── group-a-quote.md      # A. Quotes & Valuation (10 endpoints)
│   ├── group-b-financial.md  # B. Financials & Fundamentals (8 endpoints)
│   ├── group-c-capital.md    # C. Capital Flow (7 endpoints)
│   ├── group-d-limitup.md    # D. Limit-up & Sentiment (4 endpoints)
│   ├── group-e-events.md     # E. Events & Information (3 endpoints)
│   ├── group-f-sector.md     # F. Sectors & Concepts (4 endpoints)
│   ├── group-g-convertible.md # G. Convertible Bonds (2 endpoints)
│   ├── group-h-quant.md      # H. Quant & Technical (2 endpoints)
│   ├── group-i-macro.md      # I. Macro (3 endpoints)
│   ├── group-j-tools.md      # J. Tools (1 endpoint)
│   ├── group-k-agent.md      # K. Agent Enhanced (2 endpoints)
│   ├── examples.md           # Multi-endpoint analysis examples (5 scenarios)
│   └── safety-rules.md       # Financial output safety constraints (6 rules)
└── scripts/
    └── install.sh            # One-line install script

Why multi-file? A simple query like "what's the price of Moutai" only needs
the entry SKILL.md + one reference file (~4K tokens), instead of loading the
entire 47KB monolith (~15K tokens). This reduces token consumption by 70%+.

Basic Usage

All endpoints are HTTP GET, called with curl:

BASE="https://www.apocdata.com/api/blade-dataplatform/open/data"

# Get real-time quote
curl -s "$BASE/quote?symbol=000001"

# Get stock info (PE/PB/market cap)
curl -s "$BASE/stock?symbol=000001"

# Get comprehensive profile (8 dimensions in one call)
curl -s "$BASE/profile/full?symbol=688017"

Endpoint Overview

Editions & Capabilities

Apocdata is organized into four editions to match different use cases.
All editions share the same API contract; capabilities are filtered by
edition whitelist. This Skill defaults to the free-tier open endpoints.
Register to unlock higher quotas and more data capabilities.

Edition Lineup

Capability Comparison (selected)

Endpoint Verification

Last verified: 2026-08-12 12:21 CST

Verified endpoints: quote, stock, daily, financial, moneyflow, hsgt,
limit-list, announcements, concepts, macro/latest, calendar, profile/full.
Full 45-endpoint coverage is validated via the scenario quick-reference table in SKILL.md.

Notes & Compliance

• Free tier: callable without registration, suited to quick validation and light usage
• Registered users: higher quotas and more data capabilities (see the platform editions page)
• All endpoints are read-only
• Data source: Apocdata Cloud (天启云), synced with A-share market data
• Free tier quotes have a ~15-minute delay (see delayed_minutes in response)
• Announcements: T+0 at 08:00; Northbound capital: 20:00
• This Skill is for research assistance only — not investment advice
• Financial output safety constraints are enforced via references/safety-rules.md

Get Started

Free tier, callable without registration; registered users unlock higher quotas
and more data capabilities.

Try free → · Register → · Star on GitHub →

License

See LICENSE for details.