stock-analysis-mcp

English | 中文

Stock Analysis MCP + Skills

Stock Analysis MCP + Skills is a local-first A-share analysis workbench for AI Agents. It combines multiple public market-data providers into reusable local SQLite evidence: market data, indicators, sector flow, chart structure, multi-timeframe review, research reports, and reproducible backtests.

It is designed for a complete research loop: check data quality → build a candidate universe → inspect numeric and chart evidence → compare scenarios → validate rules with out-of-sample backtests. It does not provide guaranteed predictions or automatic trading instructions.

See the unified implementation plan for the data, rising-pattern, per-stock AI review, and backtest contracts.

News and financial evidence

News tools and Skills provide bounded news search, company/concept timelines, global macro context and multi-period financial metrics. Sources are official Fed/ECB RSS and Eastmoney search/financial data. Use offline=true when downloads are paused. A-share financial downloads support Sohu/Sina statements and Tencent summaries/composition, with source-specific limits. Real-time global quotes and an industry-chain knowledge graph remain outside this implementation. See source documentation.

Agent setup (npm)

npm install -g stock-analysis-mcp
stock-analysis install --agents auto   # auto-configure detected Agents + copy skills
stock-analysis setup --data-root "D:/MarketData"   # choose where SQLite data lives
stock-analysis doctor                  # verify runtime / config / agent status

Supported Agents (auto-detected and configured by install --agents auto):

Agent	Config written	Skills installed
Claude Code	~/.claude.json	✅ ~/.claude/skills/
Codex	~/.codex/config.toml	✅ ~/.codex/skills/
Cursor	~/.cursor/mcp.json	— (tools are self-describing)
VS Code Copilot	VS Code user mcp.json	—
Qoder	~/.qoder/mcp.json	✅ ~/.qoder/skills/

The installer manages an isolated Python environment with uv; install uv first if it is not already available. It asks before changing any Agent configuration and creates a backup (restorable via stock-analysis uninstall). For OpenCode and other MCP clients, a manual JSON template is printed by stock-analysis install --agents auto --dry-run.

The data directory is user-owned and is never placed in the npm package directory. Settings resolve as env var → ~/.stock-analysis/config.toml → default. Use STOCK_ANALYSIS_DATA_DIR, STOCK_ANALYSIS_STOCK_DATA_DIR, STOCK_ANALYSIS_SECTOR_DATA_DIR, and STOCK_ANALYSIS_PYTHON; legacy environment names remain supported for compatibility.

Core Capabilities

Capability	Details
Market Data	Full listed A-share universe: price / change% / PE / PB / market cap / volume ratio / turnover
K-lines	Daily OHLCV with forward/backward/no adjustment
Indicators	MA5/10/20/30/60/100/120/200/250, RSI, MACD, BOLL, KDJ, ATR%, volume ratios, returns, rolling ranges and BIAS
Popularity	Eastmoney real-time sentiment rankings + historical trends (rolling 1-year Guba rank)
Sectors	Concept and industry boards with quotes, capital flow, members, K-lines and indicators
Screening	18 composable conditions: price range / PE / PB / cap / change% / volume / turnover / sector filter
Pattern Research	Five detector families plus canonical neckline-reclaim / major-MA-rebound labels and Fibonacci confluence evidence
Deep Review	Rank 20 candidates, then inspect every monthly/weekly/daily numeric and chart packet
Backtesting	5/10/20-day event studies and next-open trading simulation with costs and out-of-sample checks

---

Quick Install

# Claude Code (recommended)
claude mcp add stock-analysis -- npx stock-analysis-mcp

# or npm global
npm install -g stock-analysis-mcp

# or pip
pip install stock-analysis-mcp

Requirements: Python >= 3.10 | Node.js >= 18

---

Data Details

Stock Market Data

Category	Fields
Basics	Latest price, open, previous close, high, low
Change	Change%, change amount, amplitude, velocity
Volume	Volume, turnover amount
Activity	Volume ratio, turnover rate
Valuation	Dynamic PE, PB, TTM PE
Market Cap	Total market cap, circulating market cap
Trend	60-day change%, YTD change%
Fund Flow	Net major capital inflow

K-line Data

Category	Details
Daily K-line	Open, high, low, close, volume, amount
Adjustment	Forward (qfq) / backward (hfq) / none
Extra Fields	Change%, turnover rate, amplitude
Identity	symbol + date + adjust_type, with source and fetch time
Compatibility	Existing databases migrate in place; legacy QFQ data is reused without a bulk copy

Technical Indicators (auto-calculated)

Category	Indicators
Moving Avg	MA5, MA10, MA20, MA30, MA60, MA100, MA120, MA200, MA250
RSI	RSI6, RSI14, RSI24
MACD	DIF, DEA, MACD histogram
Bollinger	Upper, middle, lower bands
KDJ	K, D, J values
Volatility	ATR14 and ATR%
Volume	VOL_MA5/10/20 and 5/20-day volume ratios
Returns & Range	5/10/20/60-day returns and 20/60/120-day highs/lows
Bias	BIAS20, BIAS60, BIAS250

Popularity Rankings

Category	Details
Real-time	Eastmoney popularity board (rank, price, change%, turnover rate)
History Trend	Query N-day ranking trend for any stock (AES-decrypted rolling year data)

Sector Data

Category	Details
Sector List	Concept (~400) + Industry (~80) sectors
Quotes	Index level, change%
Capital Flow	Net inflow, super-large, large, medium, small net inflow (with %)
Leaders	Sector leading stock
K-line	Historical sector index K-line (single-source Eastmoney)
Members	Member quotes, turnover, volume ratio, PE, PB

Analysis Reports

Section	Content
Overview	Price, change%, PE, PB, market cap, turnover rate
Trend	MA arrangement, 5/10/20/60-day change%, direction assessment
Indicators	RSI status, MACD golden/death cross, KDJ overbought/oversold, BOLL pos, ATR
S&R Levels	Ranked support and resistance levels from MAs and BOLL bands
Risk Assessment	Overbought/oversold, volatility, valuation, liquidity, and breakdown risk
Scenarios	Conditional confirmation, invalidation, support/resistance and risk/reward assumptions

---

MCP Tools (33 tools)

Tool	Module	Purpose
init_full_data	data_manager	quick, research (320-bar universe), or long-history full initialization; legacy quick calls remain compatible
update_daily_data	data_manager	Daily incremental refresh with `stock_kline_mode=none
get_data_status	data_manager	Local DB status (row counts / last update / paths)
screen_stocks	data_manager	Universal screening: 18 conditions + sector filter + name search + sort
get_kline_local_or_net	data_manager	Daily K-line, local-first with cached technical indicators
get_stock_kline_period	stock_data	Multi-period K-line (1/5/15/30/60m, day/week/month, real-time)
get_stock_order_book	data/auction	Timestamped five-level book; fresh pre-open observations archived locally
get_stock_auction	data/auction	Free TDX auction snapshots: price, matched/unmatched volume, measured sampling gaps; separate opening execution, not full per-second orders; precision limits
get_rank_trend_data	data_manager	Historical popularity ranking trend (N days)
get_sector_list	sector_data	Concept/industry sector list with capital flow
get_sector_hot_rank	sector_hot	THS hot concept ranking, heat, price change and signed changes between saved observations; bounded list, source as-of unknown
get_stock_belong_sectors	data_manager	Reverse lookup: stock $\rightarrow$ sectors
generate_stock_report	analysis	Technical report: trend / support & resistance / risk / position
render_stock_charts	charting	Render daily/weekly/monthly candlestick PNGs with auto trendlines (chart extra)
scan_patterns	strategies/patterns	Stock chart-pattern scan (normal or strict filter)
scan_sector_patterns	strategies/patterns	Sector chart-pattern scan (concept/industry or symbols)
get_pattern_history	strategies/patterns	Historical pattern signals for one symbol (stock/sector)
get_key_levels	strategies/patterns	Current key levels: MA system / Fibonacci / structure (highs/lows)
sync_stock_kline_universe	data/sync	Research-mode universe K-line and indicator synchronization with coverage reporting
screen_rising_candidates	tools/research	Rank the top 20 rising-pattern candidates with multi-timeframe evidence
prepare_stock_analysis	tools/research	Build the monthly/weekly/daily numeric and chart packet for per-stock AI review
find_cross_timeframe_similar_patterns	strategies/similarity	Compare the latest N bars with all historical N-bar windows and estimate conditional outcome probabilities
backtest_pattern_strategy	strategies/trading_backtest	Event study plus executable 5–20 day trading simulation
search_market_news	data/information	带来源时间、链接与缓存状态的有限新闻检索
get_stock_related_news	data/information	公司与概念消息的关联依据和时间线
get_global_market_news	data/information	官方政策发布与有限全球市场新闻
analyze_stock_financials	data/information	多期财务指标及指定同业同报告期比较
download_stock_financials	data/financials	A股多源财务下载：搜狐/新浪报表、腾讯摘要与主营构成、东方财富指标；独立缓存与分析
list_market_instruments	data/assets	Paged index, ETF/LOF, fund directory, convertible and SH/SZ exchange bond data
get_market_quote	data/assets	Confirm market/type/code in the source directory and return a delayed or cached quote
get_market_kline	data/assets	Unadjusted daily/weekly/monthly index, ETF/LOF and convertible bond K-lines
get_fund_nav	data/assets	Paged unit/accumulated NAV and subscription/redemption status
get_convertible_bond_info	data/assets	Underlying stock, conversion price/value/premium and reference trigger prices

Multi-asset tools query data on demand without stock initialization. Specify the market explicitly, e.g. sh000001 with asset_type=index; fund NAV 000001 has a separate identity. Data uses namespaced JSON caches and is not included in stock screening, pattern scans or backtests. Bond data covers exchange quotes, not interbank bonds or yields. See multi-asset examples, sources and limits.

Opening auction: price, matched and unmatched volume

Call get_stock_auction with {"symbol":"000001"} for the source's 09:15–09:25 auction-process snapshots. Each point includes the reference price, virtual matched volume, unmatched buy/sell volume and direction. Volumes are normalized to shares; amounts are calculated as price times shares. The separate opening_match contains the opening execution when available. Use get_stock_order_book for the current five-level book, or get_stock_auction with {"symbol":"000001","trade_date":"2026-10-26","offline":true} to read a saved process without networking. Stock-history initialization is not required.

The free TDX source provides sampled snapshots, not every order or every second. The response's precision reports the actual first/last timestamps, sample intervals and missing seconds. On 2026-10-26, Ping An Bank returned 67 process points with a 9-second median interval; Shanghai Pudong Development Bank returned 77 with a 6-second median. These are observations, not guaranteed refresh rates. Virtual matched volumes must not be summed, and unmatched volume is not the total order book across all prices.

Queries can run at 09:26 and after the auction. Source execution timestamps can be a few seconds after 09:25; these are retained rather than discarded. The 09:26 call path was tested with a simulated clock, and after-hours network reads succeeded, but completeness at exactly 09:26 has not been verified live. Historical availability is limited, source responses do not echo the requested date/code, and exchange-to-client latency is unverified. Provider failure falls back to clearly disclosed partial or archived data. See fields, observed precision and source limits.

Hot concept sectors: heat, price change and signed deltas

Call get_sector_hot_rank with {"limit":20,"offline":false} to retrieve the THS hot concept list. It returns heat rank, heat value, percentage price change and listing tags. The endpoint currently returns 20 concepts; this is not a full-market sector ranking.

Successful network fetches save observation snapshots. heat_delta is current heat minus the previous observed heat: positive means increased attention, negative means decreased attention. rank_improvement is previous rank minus current rank: positive means an improved rank. The response includes both observation timestamps. First observations and newly listed concepts have null deltas; leaving the list does not mean heat fell to zero. Cached reads do not create observations.

Use {"offline":true} to read cached data without networking. The cache lasts five minutes; failed requests may return explicitly marked stale data. Fetch time is not the source's market timestamp, which is unavailable, so these observations must not be described as verified live or daily changes. The source's own rank-change field is preserved as source_rank_change_raw because its comparison period and sign convention remain unverified.

THS concept codes are separate from Eastmoney BK codes and cannot be passed directly to Eastmoney sector tools. Capital flow is not automatically joined across providers. Whether high heat and strong price performance predict two further rising days still requires historical validation. See snapshot storage and source limits.

For fixed-time research, schedule python -m stock_analysis_mcp.cli sector-hot-1430 --capture at 14:30 Asia/Shanghai. It accepts fresh observations only within [14:30,14:35) on verified trading days and preserves the first successful capture per day. Read with get_sector_hot_rank(comparison="daily_1430", trade_date="2026-10-21") or CLI sector-hot-1430 --date 2026-10-21. Comparisons use the exact previous trading day's capture, including signed heat changes, percentage changes and rank improvements; missing captures stay missing. The actual observation time is retained, and source freshness remains unverified. The CLI does not install a scheduler or reconstruct historical intraday lists.

Search ETFs and retrieve K-lines through a Skill

The stock-analysis Skill includes multi-asset routing. For example, ask your agent:

Find STAR 50 ETFs, list their codes and names, then retrieve year-to-date daily K-lines for ChinaAMC STAR 50 ETF (588000).

The agent calls list_market_instruments(asset_type="etf", page=1, page_size=100), follows has_more, filters names, and confirms the market/type/code identity. There is no dedicated keyword search parameter yet. STAR 50, STAR & ChiNext 50, and STAR 50 enhanced products should be separated. If pagination fails, report incomplete coverage rather than presenting a partial list as complete.

After confirming sh588000, call get_market_kline with these arguments:

{
  "symbol": "sh588000",
  "asset_type": "etf",
  "start_date": "2026-01-01",
  "period": "daily",
  "limit": 320
}

period supports daily, weekly, and monthly. The tool returns at most limit unadjusted bars within the requested date range. Adjust start_date for another year; omitting end_date uses today's date in Asia/Shanghai. Use get_market_quote for quotes and get_fund_nav for fund NAV. Ordinary bonds currently support lists and quotes only.

These on-demand queries do not require stock database initialization. offline=true reads only the matching request cache; results disclose cache misses, unavailable sources, and stale data. ETF K-lines are not yet integrated with the existing stock chart renderer, pattern scanners, or backtests.

After updating source, restart the MCP service that uses this project's entrypoint. If your agent uses a previously installed Skill copy, also update its stock-analysis/SKILL.md; editing repository files does not automatically replace installed copies.

MCP call contract

Every tool publishes a JSON Schema. The server validates required and unknown fields, basic types, enum values, and numeric ranges. Results use {data, meta, warnings, error}: check error first, preserve warnings, and use both meta and the data cutoff date for freshness. The server does not provide persistent monitoring, push notifications, or deterministic buy/sell decisions.

Ships with 14 Agent Skills (installed automatically by stock-analysis install):

Skill	Purpose
stock-analysis	Main router — stock research readiness plus index, ETF/LOF, fund NAV and bond queries
stock-analysis-data-init	Existing-data reuse, initialization, coverage-aware updates and troubleshooting
stock-analysis-stock-screening	Composing screening conditions and workflows
stock-analysis-report-generation	Evidence-based single-stock report and conditional scenarios
stock-analysis-multi-timeframe	Monthly/weekly/daily/intraday analysis and conflict resolution
stock-analysis-strategy-backtest	Chart pattern backtesting and parameter optimization
stock-analysis-chart-trend	Per-stock visual structure attribution with pivot and line uncertainty checks
stock-analysis-rising-patterns	Rank 20 candidates, deeply review every monthly/weekly/daily chart, and summarize common traits
a-share-dividends	Historical/latest dividends and announced future dates, with source and coverage checks
stock-analysis-st-removal	Forward-looking ordinary-ST removal research: current universe, trigger timeline, remaining conditions and evidence-backed windows

Ordinary-ST removal research

stock-analysis-st-removal matches each company's active warning causes to board-specific removal rules. Reports show the ST start date, actual cause, required evidence, removal thesis, remaining barriers and conditional window. Pre-application, submitted, approved-but-not-effective and effective events are distinguished. Profit/loss is an independent investment-risk filter, not a universal legal removal gate; the audited-profit or resolved-going-concern route is preserved where applicable.

The 109-company example dated 2026-10-18 includes HTML, CSV, JSON and rule notes. It correctly separates ST Juewei's disclosed removal decision (effective October 20) from pending applications. HTML supports window/evidence sorting and stage/profit filters while displaying warning dates and causes directly. This is a research snapshot, not a live recommendation or point-in-time backtest dataset.

python -X utf8 skills/stock-analysis-st-removal/scripts/st_research_collect.py --output <new-report-directory> --as-of YYYY-MM-DD
# Review original filings and write reviewed_notes.json before rendering.
python -X utf8 skills/stock-analysis-st-removal/scripts/st_research_report.py <report-directory> --as-of YYYY-MM-DD --overrides <report-directory>/reviewed_notes.json
python -X utf8 skills/stock-analysis-st-removal/scripts/st_research_refresh.py <report-directory> --project . --as-of YYYY-MM-DD --since YYYY-MM-DD

Replace the placeholders with actual dates/paths and overlap --since with the previous successful check. The cause-specific rule reference links official sources. Availability filtering uses actual publication time while retaining the printed notice date separately. Rendering is offline; refresh evidence and review conclusions before regenerating a current report. These helpers do not add a registered MCP tool.

A-share dividends

This project owns a-share-dividends, migrated from the sibling trading repository. Its standalone script queries Eastmoney by symbol or report period, checks pagination and archives raw JSON. Formal-announcement verification is an agent workflow; payment dates, full-market latest-event coverage and account adjustments are not automatically completed by this script. It is a Skill helper, not an additional registered MCP tool.

python -X utf8 skills/a-share-dividends/scripts/fetch_dividends.py --symbol 000001 --as-of 2026-10-09 --output <output-dir>/000001-dividends-20261009.json

Use an existing output directory and a new filename. Source limits and field definitions are in references/sources.md.

What each Skill delivers

The Skills are output-oriented: each one turns MCP tool results into a consistent, reviewable artifact rather than a loose paragraph.

Skill	Result view	Typical result contents
stock-analysis	Research route card	Chosen workflow, data-quality gate, tool chain, warnings, and next action
stock-analysis-data-init	Data health report	Expected trading date, latest date, per-dataset coverage, successes/failures, provider warnings, remaining gaps
stock-analysis-stock-screening	Sortable candidate table	Code/name, cutoff, filters, pattern/stage, score components, sector context, support/resistance/invalidation, rejected or missing-data notes
stock-analysis-report-generation	Single-stock evidence report	Conclusion/confidence, data block, monthly/weekly/daily evidence, key-level table, scenarios, risks, limitations
stock-analysis-multi-timeframe	Timeframe comparison matrix	Monthly/weekly/daily/intraday metrics, agreement/conflict matrix, primary/alternate view, exact confirmation and invalidation levels
stock-analysis-chart-trend	Chart review card	Chart cutoff, pivots, trendlines/channels/ranges/W-M/Fibonacci zones, program-vs-chart agreement, volume evidence
stock-analysis-rising-patterns	Candidate cards + comparison	One evidence card per returned stock, reviewed ranking, current-vs-history commonality, and coverage caveats
stock-analysis-strategy-backtest	Backtest result pack	Universe/period, coverage, signal/trade counts, assumptions, event and portfolio metrics, train/test and yearly stability, report/CSV paths
stock-analysis-st-removal	Cause-specific removal evidence table	ST date/cause, board rules, removal thesis, missing evidence, event stage, profit/loss risks, sources and freshness

Example compact result card:

000001 平安银行 · 2026-09-17 · qfq · 486 daily bars
Structure: weekly pullback / daily neckline reclaim (candidate → triggered)
Evidence: close 12.34 | MA20 12.10 | neckline 12.28 | volume 1.42x
Confirm: daily close above 12.28 and hold; Invalidate: close below 11.86
Warnings: sector data delayed; chart confirmation available

Exact values, dates, coverage and warnings always come from the current MCP response; the example above only illustrates the presentation format.

---

Research workflow

get_data_status
  → update or fill only missing coverage
  → screen_rising_candidates(top_n=20)
  → prepare_stock_analysis for every returned stock
  → inspect monthly, weekly and daily charts
  → compare reviewed candidates and extract common traits
  → backtest_pattern_strategy(mode="both")
  → human approval before any production-rule change

A full-market claim requires at least 95% eligible-stock K-line coverage. The five low-level pattern detectors are trend_pullback, w_bottom, m_neckline, box_breakout, and ma_rebound. High-level reports rename two of them to neckline_reclaim and major_ma_rebound; fibonacci_confluence is supporting evidence rather than a standalone reversal detector.

Unified CLI

In addition to the MCP server, a unified CLI is provided for database maintenance, backfilling, pattern analysis, and strategy backtesting:

# Rebuild local databases with dry-run preview
python -m stock_analysis_mcp.cli rebuild --dry-run
python -m stock_analysis_mcp.cli rebuild --workers 8 --with-sectors

# Auto-detect and backfill historical data gaps
python -m stock_analysis_mcp.cli backfill --start 2026-01-01

# Daily evening capture after market close
python -m stock_analysis_mcp.cli daily-capture

# Cleanup database redundancy and VACUUM
python -m stock_analysis_mcp.cli cleanup

# Pattern scanning and backtesting
python -m stock_analysis_mcp.cli pattern-scan --universe sectors --date 2026-08-14
python -m stock_analysis_mcp.cli pattern-backtest --universe stocks --sample 300
python -m stock_analysis_mcp.cli pattern-optimize --cache signals.csv

---

Local Database

Data is stored in local SQLite databases using WAL mode for high-concurrency read/write operations:

Database	Default Path (Windows)	Default Path (Linux/macOS)	Contents
Stock DB	~/Desktop/股票信息/stock_data.db	~/.stock-analysis/data/stocks/stock_data.db	stock quotes + adjustment-aware K-lines + popularity + indicators + combined + coverage/signals
Sector DB	~/Desktop/分析板块/sector_data.db	~/.stock-analysis/data/sectors/sector_data.db	sector quotes + capital flow + Eastmoney-only K-lines + members + indicators

Override with STOCK_ANALYSIS_STOCK_DATA_DIR and STOCK_ANALYSIS_SECTOR_DATA_DIR.

---

Environment Variables

Resolution order: env var → ~/.stock-analysis/config.toml → default.

Variable	Default	Purpose
STOCK_ANALYSIS_PYTHON	managed runtime $\rightarrow$ python	Python interpreter for the Node shim
STOCK_ANALYSIS_DATA_DIR	Win: ~/Desktop; Linux/macOS: ~/.stock-analysis/data	Root directory for both databases
STOCK_ANALYSIS_STOCK_DATA_DIR	<data_root>/股票信息 (Linux/macOS: stocks)	Stock database directory
STOCK_ANALYSIS_SECTOR_DATA_DIR	<data_root>/分析板块 (Linux/macOS: sectors)	Sector database directory
STOCK_ANALYSIS_CONFIG	~/.stock-analysis/config.toml	Override config.toml path
EASTMONEY_COOKIE	auto-extract from Edge	Manual Eastmoney API cookie string

---

Client Configuration

Claude Desktop / Claude Code

claude mcp add stock-analysis -- npx stock-analysis-mcp

Or manually in claude_desktop_config.json:

{
  "mcpServers": {
    "stock-analysis": {
      "command": "npx",
      "args": ["stock-analysis-mcp"]
    }
  }
}

Codex

Run stock-analysis install --agents codex, or add to ~/.codex/config.toml:

[mcp_servers.stock-analysis]
command = "npx"
args = ["stock-analysis-mcp"]

Cursor

Run stock-analysis install --agents cursor, or add the JSON block to ~/.cursor/mcp.json.

VS Code Copilot

Run stock-analysis install --agents copilot, or add to VS Code user mcp.json (Windows: %APPDATA%\Code\User\mcp.json):

{
  "servers": {
    "stock-analysis": {
      "type": "stdio",
      "command": "npx",
      "args": ["stock-analysis-mcp"]
    }
  }
}

Qoder

Run stock-analysis install --agents qoder, or add to ~/.qoder/mcp.json.

OpenCode / Other MCP Clients

{
  "mcpServers": {
    "stock-analysis": {
      "command": "npx",
      "args": ["stock-analysis-mcp"]
    }
  }
}

Direct Python Execution

{
  "mcpServers": {
    "stock-analysis": {
      "command": "python",
      "args": ["-m", "stock_analysis_mcp.server"]
    }
  }
}

---

Example Usage

# Reuse an existing DB; initialize only when required
init_full_data(mode="quick")

# Recommended before full-market rising-pattern screening
init_full_data(mode="research", workers=6, resume=True)

# Or long-history full init (large resumable download for backtesting)
init_full_data(mode="full", workers=6, resume=True)

# Daily update after market close; compare freshness with the latest trading day
update_daily_data(stock_kline_mode="tracked")

# 3. Screen: change > 3%, PE < 30, volume ratio > 1.5
screen_stocks({"min_change_pct": 3, "max_pe": 30, "min_volume_ratio": 1.5})

# 4. Screen: semiconductor sector with PE < 50, sorted by popularity rank
screen_stocks({"max_pe": 50}, sector_code="BK1090", sort_by="popularity_rank")

# 5. Search: stocks containing "银行" (Bank)
screen_stocks(name_keyword="银行")

# 6. Get Ping An Bank K-line with cached indicators
get_kline_local_or_net("000001", days=250)

# 7. Ping An Bank 30-day popularity trend
get_rank_trend_data("000001", days=30)

# 8. Generate full analysis report
generate_stock_report("000001")

# 9. Chart pattern scan: market-wide strictly filtered signals
scan_patterns(strict=True)

# 10. Rank 20 rising candidates, then prepare every candidate for AI chart review
screen_rising_candidates(top_n=20, strict=True)
prepare_stock_analysis("000001", days=500, include_chart=True)

# 11. Validate event outcomes and executable portfolio assumptions
backtest_pattern_strategy(mode="both", split="2022-01-01")

---

Development

git clone https://github.com/lalal-zzz/stock-analysis-mcp.git
cd stock-analysis-mcp
pip install -e ".[dev]"
pytest                    # Python unit tests
npm run test:node         # Node installer tests
pytest -m integration     # Opt-in: real network + local DB writes

Data consistency and maintenance

See data repair and validation for read-only audits, archived full-history repair, resumable retries, local zone alerts, reviewed rule activation, and daily-equity backtest outputs. Historical row coverage, current trading-day coverage, and verified price-basis coverage are reported separately. The installer includes chart dependencies. CI tests both supported MCP SDK generations before publishing.

License

MIT License

News and financial Skills

Skill	Purpose
stock-analysis-news	News evidence and citations
stock-analysis-related-news	Company/concept event timelines
stock-analysis-global-markets	Macro policy and global market context
stock-analysis-financials	Multi-period financial metrics and peers