tradeos-mcp
Connects TradeOS trading intelligence to MCP-compatible AI tools, enabling natural language-driven market analysis, ticker search, spread comparison, macro context, and custom trading agent management.
README
tradeos-mcp
TradeOS MCP lets you connect TradeOS trading intelligence to MCP-compatible AI tools, so your assistant can analyze markets, search tickers, compare spreads, use macro context, and manage custom trading agents directly from natural-language prompts.
Overview
This repo publishes @tradeos/tradeos-mcp on npm: a stdio MCP bridge that proxies tools to TradeOS Streamable HTTP (mcp-call). You can connect in two ways:
| Mode | Best for | Auth |
|---|---|---|
| HTTP (recommended) | Cursor, Claude Code, most MCP clients | Browser OAuth in the client |
| npm stdio | Clients that only support command / stdio |
TRADEOS_ACCESS_TOKEN env var |
Production endpoint (both modes ultimately call this):
https://ai.tradeos.xyz/api/agent/mcp/mcp-call
Product docs: TradeOS MCP integration & usage
Prerequisites
- Node.js 18+
- A TradeOS account with MCP API access
- For npm stdio: ability to run a local install (see Windows notes)
Option A — HTTP (recommended)
No npm install. The client handles OAuth.
Cursor
Edit user config ~/.cursor/mcp.json (Windows: %USERPROFILE%\.cursor\mcp.json):
{
"mcpServers": {
"tradeos": {
"url": "https://ai.tradeos.xyz/api/agent/mcp/mcp-call"
}
}
}
Save, restart Cursor (or refresh MCP in Settings → MCP), then complete OAuth when prompted.
Claude Code (plugin — recommended)
Use the bundled plugin in tradeos-skills/:
claude --plugin-dir ./tradeos-skills
Then complete OAuth and run /tradeos:analyze. Full plugin workflow (validate, community marketplace submit): tradeos-skills/README.md — aligned with Claude plugin docs.
Option B — npm stdio bridge
Use when your client requires a local process (command + stdio).
1. Install locally
Do not rely on npx inside MCP on Windows (cache/path issues). Install once to a fixed directory:
mkdir -p ~/tradeos-mcp-run # Windows: mkdir %USERPROFILE%\Downloads\tradeos-mcp-run
cd ~/tradeos-mcp-run # Windows: cd %USERPROFILE%\Downloads\tradeos-mcp-run
npm install @tradeos/tradeos-mcp
Pin a version if you prefer, e.g. @tradeos/tradeos-mcp@1.0.0.
2. Get TRADEOS_ACCESS_TOKEN
Run the OAuth helper (opens browser → log in on TradeOS → prints token):
# Linux / macOS
npx -y -p @tradeos/tradeos-mcp tradeos-mcp-oauth
# Or subcommand on main bin:
npx -y @tradeos/tradeos-mcp oauth
# Or, from your install directory:
node node_modules/@tradeos/tradeos-mcp/scripts/fetch-token.mjs
Windows (recommended):
cd /d %USERPROFILE%\Downloads\tradeos-mcp-run
node node_modules\@tradeos\tradeos-mcp\scripts\fetch-token.mjs
Copy the printed JWT (eyJ...). For local development of this repo, copy into .env:
cp .example.env .env
# TRADEOS_ACCESS_TOKEN=eyJ...
End users who only npm install the package do not need .env — set the token in MCP client config (mcp.json) instead.
Do not commit tokens to git.
3. Configure Cursor
{
"mcpServers": {
"tradeos-npm": {
"command": "node",
"args": [
"C:/Users/YOU/Downloads/tradeos-mcp-run/node_modules/@tradeos/tradeos-mcp/build/index.js"
],
"env": {
"TRADEOS_ACCESS_TOKEN": "eyJ..."
}
}
}
}
Replace YOU and the path with your actual install location. Use forward slashes or escaped backslashes in JSON.
If Cursor cannot find node, set the full path:
"command": "C:/nvm4w/nodejs/node.exe"
Enable tradeos-npm in Settings → MCP. Status should show connected with tools listed.
4. Verify in terminal (optional)
Environment variables in mcp.json do not apply to your shell. Set the token in the same terminal session:
REM Windows CMD
set TRADEOS_ACCESS_TOKEN=eyJ...
node C:\Users\YOU\Downloads\tradeos-mcp-run\node_modules\@tradeos\tradeos-mcp\build\index.js
# Linux / macOS
export TRADEOS_ACCESS_TOKEN=eyJ...
node ~/tradeos-mcp-run/node_modules/@tradeos/tradeos-mcp/build/index.js
Success:
[tradeos-mcp] remote: https://ai.tradeos.xyz/api/agent/mcp/mcp-call
[tradeos-mcp] stdio bridge ready
Press Ctrl+C to stop. Cursor starts its own process; you do not need to keep this terminal open.
Windows notes
- Use local
node+ install path inmcp.json, notnpx, for MCP. - If npm cache errors mention
C:\Program Files\nodejs\node_cache, run once:npm config set cache "%USERPROFILE%\.npm-cache" npx @pkg tradeos-mcp-oauthmay run the wrong bin on Windows. Prefer:
ornpx -y -p @tradeos/tradeos-mcp tradeos-mcp-oauthnode .../scripts/fetch-token.mjsfrom your install directory.- Clear a broken npx cache (quit Cursor first):
rmdir /s /q "%USERPROFILE%\.npm-cache\_npx"
npm package reference
| Item | Value |
|---|---|
| Package | @tradeos/tradeos-mcp |
| MCP Registry name | io.github.TradeOS-AI/tradeos-mcp |
| Bin: MCP bridge | tradeos-mcp |
| Bin: OAuth token | tradeos-mcp-oauth |
| OAuth subcommand | oauth on bin tradeos-mcp |
Install:
npm install @tradeos/tradeos-mcp
Development (this repo)
git clone https://github.com/TradeOS-AI/tradeos-mcp.git
cd tradeos-mcp
npm ci
npm run build
npm run oauth:token # browser OAuth → prints token (copy into .env if needed)
npm run registry:verify
| Script | Purpose |
|---|---|
npm run build |
Compile stdio bridge to build/ |
npm run oauth:token |
Browser OAuth → print token |
npm run registry:verify |
Pre-publish checks (package.json ↔ server.json) |
npm run registry:sync-version <ver> |
Sync version from release tag |
Publishing (maintainers)
Push a version tag to trigger .github/workflows/publish-mcp.yml:
git tag v1.0.0
git push origin v1.0.0
- Stable tags (
v1.0.0) publish to npmlatestand the MCP Registry. - Pre-release tags (
v1.0.0-beta.1) publish with an npm dist-tag matching the prerelease label (e.g.beta). - Requires GitHub Actions secret
NPM_TOKEN(Classic Automation token for@tradeos).
Verify on the registry API:
curl "https://registry.modelcontextprotocol.io/v0.1/servers?search=io.github.TradeOS-AI/tradeos-mcp"
Troubleshooting
| Symptom | Likely cause | Fix |
|---|---|---|
TRADEOS_ACCESS_TOKEN is required |
Token not set in env / mcp.json |
Run oauth helper; set env in MCP config |
unauthorized on connect |
Expired or invalid token | Re-run oauth; paste fresh JWT (no Bearer prefix) |
EPERM / node_cache |
npm cache under Program Files |
npm config set cache "%USERPROFILE%\.npm-cache" |
TAR_ENTRY_ERROR / 'tradeos-mcp' is not recognized |
npx failed on Windows |
Local install + node path in mcp.json |
MODULE_NOT_FOUND for node_modules/... |
Ran script from wrong directory | cd to install dir or use absolute path |
| Token works in terminal, not in Cursor | mcp.json path or token mismatch |
Match paths; restart Cursor / refresh MCP |
| MCP connects but no tools | Wrong server enabled | Disable broken localhost entries; use production HTTP URL or working tradeos-npm |
Related
tradeos-skills/— Claude Code plugin + analyze skillserver.json— MCP Registry manifest- TradeOS MCP FAQ (GitBook)
推荐服务器
Baidu Map
百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。
Playwright MCP Server
一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。
Magic Component Platform (MCP)
一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。
Audiense Insights MCP Server
通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。
VeyraX
一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。
graphlit-mcp-server
模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。
Kagi MCP Server
一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。
e2b-mcp-server
使用 MCP 通过 e2b 运行代码。
Neon MCP Server
用于与 Neon 管理 API 和数据库交互的 MCP 服务器
Exa MCP Server
模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。