Finance Toolkit
500+ transparent financial methods calculated from raw statements, not third-party endpoints.
电脑或云端
Agent Trading Journal
概览
作者提供的说明（英文原文）
AI trading journal: your agent logs trades from chart screenshots and checks your own rules.
PopNx MCP 的三大支柱
这三项永远不会合并成一个分数：意见、测量和人气是不同的东西。
- 用户评价i暂无评价评价功能将在后续阶段与账户功能一起开放
- 我们的基础检查i1 项中通过 1 项依据公开信息检查；尚未扫描代码
- 使用情况i非常新↓ 63/周 · ★ –
它能帮你做什么
我们还没有为它写通俗的介绍。下面是我们从注册中心读取到的事实。
- 连接方式: 在你的电脑上运行
- 传输方式: stdio
- 分类: 会计、财务与税务
安装
选择你使用的 AI 应用，然后复制设置。
安装之前
- 基础检查：1 项中通过 1 项
- 在你的电脑上运行，拥有你的用户账户的权限（文件、网络、应用）。
- 需要你自己填写：
JOURNAL_DATA_DIR
你需要
- npm已安装 Node.js（用于 npx 命令）
你用的是哪个应用？
在上方选择你的应用。我们会记住它，下次按钮会显示“安装到 …”。
打开 Claude Desktop → Settings → Developer → Edit Config，把下面的内容粘贴到 claude_desktop_config.json
{
"mcpServers": {
"agent-trading-journal": {
"command": "npx",
"args": [
"-y",
"agent-trading-journal"
],
"env": {
"JOURNAL_DATA_DIR": "your-value"
}
}
}
}在终端中运行这条命令
claude mcp add agent-trading-journal --env JOURNAL_DATA_DIR=your-value -- npx -y agent-trading-journal运行前请先填写空着的值。
或者自己复制设置
粘贴到项目中的 .cursor/mcp.json，或 ~/.cursor/mcp.json
{
"mcpServers": {
"agent-trading-journal": {
"command": "npx",
"args": [
"-y",
"agent-trading-journal"
],
"env": {
"JOURNAL_DATA_DIR": "your-value"
}
}
}
}浏览器会先询问是否打开应用，然后应用会显示设置并请你确认。密钥的值会留空，由你自己填写。
或者自己复制设置
粘贴到项目中的 .vscode/mcp.json
{
"servers": {
"agent-trading-journal": {
"type": "stdio",
"command": "npx",
"args": [
"-y",
"agent-trading-journal"
],
"env": {
"JOURNAL_DATA_DIR": "your-value"
}
}
}
}浏览器会先询问是否打开应用，然后应用会显示设置并请你确认。密钥的值会留空，由你自己填写。
这个服务器在你的电脑上运行，而 ChatGPT 只能连接互联网上的服务器。请选择其他应用。
你的选择通过 Cookie 保存在这台设备上，无需账户。
需要你填写的值
JOURNAL_DATA_DIRenvFolder for journal.db and screenshots (default ~/.agent-trading-journal)
这些设置是根据作者在注册中心的条目生成的。如果 README 的写法不同，请以 README 为准。
它能做什么
它在你的电脑上运行；安装后，工具列表会显示在你的 AI 应用中。
基础检查i
我们检查的是任何人都能看到的信息；我们没有阅读代码。不适用的检查项不计入总数。 · 检查于 2026年11月9日
- ✓ 通过
作者与仓库一致
注册名称已通过 GitHub 账户 smizxe 验证，该账户拥有此仓库
- … 尚未检查
仍在维护
我们还没有从 GitHub 读取这个仓库
- … 尚未检查
有许可证
我们还没有从 GitHub 读取这个仓库
- – 不适用
端点有响应
它在你的电脑上运行，没有互联网端点
- – 不适用
使用 OAuth 登录
在你的电脑上运行，改用本地设置
- … 尚未检查
工具描述中没有隐藏指令
它在你的电脑上运行；我们只能读取互联网端点的工具列表
与“会计、财务与税务”中的其他服务器比较i
- 星标数多于此分类中其他 2964 个服务器的 0%
- 发布到注册中心的时间比此分类中其他 2964 个服务器的 41% 更新
- 基础检查结果等于或优于此分类中其他 2964 个服务器的 100%
与注册中心里这个分类中所有活跃的服务器比较。只有当发布者拥有该仓库时，星标数才计入。
常见问题
Agent Trading Journal 安全吗？
它在 1 项适用的基础检查中通过了 1 项。我们没有阅读代码。在添加密钥或令牌之前，请先看看这个服务器要求哪些权限，并使用只具备所需权限的密钥。
它需要哪些密钥或代码？
注册中心没有列出必需的密钥，不过有些服务器会在首次使用时要求你登录。
它在我的电脑上运行还是在云端？
它在你的电脑上运行。你的 AI 应用会在需要时启动它，数据不会经过作者的服务器（除非它自己调用在线服务）。
它还在维护吗？
GitHub 仓库最后更新于–，最新版本（0.1.0）于2个月前进入注册中心。
哪些应用可以使用它？
支持 MCP 的应用，例如 Claude Desktop、Claude Code、Cursor 和 VS Code。你的电脑上还需要有 Node.js。
有什么可以替代它？
同一分类中有：Finance Toolkit, SolidInvoice, Korea Tax Calculator
替代选择
同一分类中其他作者的服务器，按星标数排序。
如何看卡片：↓ 每周下载量i★ GitHub 星标i三大支柱：评价 · 检查 · 使用情况i更新 = 最新版本发布到注册中心的时间i
SolidInvoice
Create and send invoices and quotes, manage clients, and track payments in SolidInvoice.
在云端运行✓ solidinvoice.co
Korea Tax Calculator
Korean tax calculation and tax-audit judgment MCP: 96 tools with statutory citations.
在你的电脑上运行
作者的 README
打开 README
按原文从作者的仓库复制。PopNx MCP 没有审查或背书其内容。HTML、图片和脚本已被移除。链接会离开本站。
Agent Trading Journal: an AI trading journal (MCP server) for Claude Code and Codex
Works with: Claude Code · Codex CLI · Claude Desktop · Cursor · Windsurf · any MCP client — Markets: forex, futures, stocks, crypto — Imports: MetaTrader 4/5, TradingView, IBKR, NinjaTrader, Tradovate and more
A trading journal your AI agent writes for you. Paste a chart screenshot into Claude Code, Codex or any MCP client. The agent reads it, fills in *your* strategy's fields, checks the trade against *your* rules, logs it, and attaches the image. A local dashboard shows what's working.
No forms, no LLM API key, no account. Everything stays on your machine in one SQLite file.
Dashboard with generated demo data (npx agent-trading-journal demo). Not a real account.
Why this exists
Most "AI trading journals" are a web form plus a chatbot that reads your numbers afterwards. Two problems follow:
- Logging is the bottleneck. People stop journaling because entering trades by hand is tedious.
- Generic fields measure nothing specific. "Setup: A/B/C" doesn't tell you whether *your* rules work.
Here the agent is the data-entry clerk, and the journal is shaped by an onboarding interview about how *you* trade:
| Typical AI journal | Agent Trading Journal | |
|---|---|---|
| Who enters trades | You, in a form | Your agent, from screenshots (or a broker statement) |
| How AI connects | The app calls an LLM with your API key | MCP: your agent (Claude Code / Codex / …) *is* the AI |
| What is measured | Fixed fields + free tags | Fields and a rule checklist defined for your strategy |
| Rules | Notes | Each rule is pass/fail per trade → "E[R] when followed vs broken" |
| Skipped setups | Not tracked | Logged and reviewed (was staying out right?) |
| Risk limits | Rarely | Prop-firm style guard: daily loss, max drawdown, trades/day, loss streak |
Quick start
1. Connect it to your agent. Requires Node.js 22+.
Claude Code:
claude mcp add journal -- npx -y agent-trading-journalCodex CLI (~/.codex/config.toml):
[mcp_servers.journal]command = "npx" args = ["-y", "agent-trading-journal"]
Claude Desktop, Cursor, or any other MCP client: add a stdio server with command npx and args ["-y", "agent-trading-journal"]. Run npx agent-trading-journal setup to print all the snippets.
2. Tell your agent: *"Set up my trading journal."*
It interviews you in your language: markets, risk rules, how you find bias, setups, entries, stops and targets, and when you don't trade. From your answers it proposes fields (what to measure on every trade) and rules (your entry checklist). You confirm, and it saves them. See docs/ONBOARDING.md for the full script.
3. Trade or backtest as usual. Paste screenshots and say *"log this"*. Ask for *"weekly review"* or *"pre-session check"*. Say *"open the dashboard"* to see it at http://localhost:3777.
Just looking? npx agent-trading-journal demo fills a separate demo journal and opens the dashboard.
What the agent can do (MCP tools)
| Area | Tools |
|---|---|
| Setup | get_setup_status, get_onboarding_guide, set_profile, list_presets, import_preset, upsert_strategy, get_strategy, list_strategies, export_strategy |
| Accounts & risk | upsert_account, list_accounts, check_risk |
| Journal | start_session, list_sessions, log_trade, update_trade, delete_trade, query_trades, get_trade, add_screenshot |
| Lessons | add_lesson, update_lesson, delete_lesson, list_lessons |
| Analysis | get_stats, run_sql_readonly, export_review (Markdown, Obsidian-friendly) |
| Import | import_statement: MetaTrader 4/5, TradingView, IBKR, ThinkorSwim, NinjaTrader, Tradovate, TopstepX, Webull, DAS, TradeZella, Tradervue, generic CSV |
| Dashboard | start_dashboard, stop_dashboard, dashboard_status |
MCP prompts (slash commands in clients that support them): onboarding, log_trade_from_screenshot, weekly_review, pre_session_check. The same workflows are available through the get_workflow tool for clients without prompt support.
Concepts
- Strategy: your plan (bias, setup, entry, stop, targets, management, no-trade conditions), plus:
- Fields: per-trade variables you want to analyse (
enum,bool,number,text). Every non-text field gets its own breakdown chart. - Rules: your checklist, each
mustorshould. Rule adherence compares expectancy when all must-rules were followed against when at least one was broken. - Versions: when fields or rules change, the version bumps. Old trades keep the version they were logged under, and removed fields are archived, not deleted.
- Session: a backtest, forward-test or live block on one symbol, optionally tied to an account.
- R: result in multiples of planned risk. Stats are R-based, so they compare across instruments and account sizes.
- Presets:
presets/has a *starter* template (for traders without a written strategy), a trend pullback and an opening-range breakout. They're starting points to edit, not recommendations. - Edge Score: an open 0–100 composite (docs/edge-score.md), withheld under 5 trades.
Data and configuration
| Env var | Default | Purpose |
|---|---|---|
JOURNAL_DATA_DIR | ~/.agent-trading-journal | folder with journal.db + screenshots/ |
JOURNAL_DB, JOURNAL_SCREENSHOTS | inside the data dir | override either path |
JOURNAL_PORT | 3777 | dashboard port (binds to 127.0.0.1 only) |
JOURNAL_WEB | unset | 1 = start the dashboard together with the MCP server |
JOURNAL_LOCALES_DIR | /locales | extra dashboard languages as .json (docs/locales.md) |
Back up by copying the data dir. run_sql_readonly and the SQLite file are yours to query.
CLI
agent-trading-journal [mcp] MCP server on stdio (what MCP clients run)agent-trading-journal dashboard dashboard at http://localhost:3777 agent-trading-journal demo demo data in a separate dir + dashboard agent-trading-journal import --account [--strategy ] [--dry-run] agent-trading-journal setup config snippets for Claude Code / Claude Desktop / Codex
Safety
This is a journaling and analysis tool. It does not place trades, connect to brokers for execution, or give financial advice. Leveraged trading is high-risk. The agent's chart reading can be wrong: every logged value is visible and editable, and rule checks show which ones failed and why.
FAQ
Do I need an OpenAI or Anthropic API key?
No. The journal is an MCP server; the AI is the agent you already use (Claude Code, Codex, Claude Desktop, Cursor…). The journal itself never calls an LLM.
Where is my data stored? Is anything sent online?
Everything is in one local SQLite file plus a screenshots folder (~/.agent-trading-journal by default). The dashboard binds to 127.0.0.1. Nothing is uploaded anywhere by the journal.
Can I use my own trading strategy?
That's the point. The onboarding interview turns *your* plan into custom fields and a rule checklist. Presets are only starting points. SMC/ICT, price action, supply & demand, indicators, options: anything you can describe works.
Can I import trades from MetaTrader 5 or TradingView?
Yes: import_statement (or agent-trading-journal import --account ) reads MT4/MT5 statements, TradingView paper/strategy exports, IBKR, ThinkorSwim, NinjaTrader, Tradovate, TopstepX, Webull, DAS, TradeZella and Tradervue exports. Your agent can then add screenshots and rule checks to the imported trades.
Does it work for backtesting (TradingView Replay) and prop firm challenges?
Yes. Sessions are backtest, forward or live. Accounts can carry prop-firm limits (daily loss, max drawdown, trades per day, loss streak), and check_risk warns before you hit them.
Does it place trades or give signals?
No. It only records and analyses. You make every trading decision.
Credits and license
MIT © 2026 smizxe.
Statement parsing and round-trip reconstruction use @luxalgo/journal-importers and @luxalgo/journal-core (MIT) from LuxAlgo's Trade Journal. The Edge Score is adapted from their open formula. This project is not affiliated with or endorsed by LuxAlgo.