B
blink1217/trading-swarm-mcp
Free offline zero-trust backtest forensics (lookahead bias, Newey-West + BH t-stats, cost survival) and…
更新 2026-11-19
在 GitHub 查看 ↗直链 · 无中转页
安装与配置
远端端点(Claude Desktop / Cursor / Windsurf 共用 mcpServers)
{
"mcpServers": {
"blink1217-trading-swarm-mcp": {
"url": "https://swarm-mcp-503318750546.europe-west1.run.app/mcp/data"
}
}
}把这段粘进 Claude Desktop / Cursor 的 mcpServers 配置即可。端点来自公开证据,我们没有替你安装或托管它。 官方 Registry 记录:这个端点列了请求头 Authorization(Bearer <token from https://1.21initiative.com/mcp/> (or append ?apiToken=<token> to the URL).),但没有标为必填。能不能匿名连上,以项目文档为准。
VS Code(.vscode/mcp.json —— 键名是 servers,远程要 type: http)
{
"servers": {
"blink1217-trading-swarm-mcp": {
"type": "http",
"url": "https://swarm-mcp-503318750546.europe-west1.run.app/mcp/data"
}
}
}VS Code 的配置文件和另外三家不同:顶层键是 <code>servers</code>,远程端点还要写 <code>"type": "http"</code>。复制到 .vscode/mcp.json。
Claude Code(命令行)
claude mcp add --transport http blink1217-trading-swarm-mcp https://swarm-mcp-503318750546.europe-west1.run.app/mcp/data
Claude Code 不收 JSON 文件,用这条命令直接加。
配置文件在:macOS
~/Library/Application Support/Claude/claude_desktop_config.json、Windows %APPDATA%\Claude\claude_desktop_config.json、Linux ~/.config/Claude/claude_desktop_config.json。配置项的键名以你的客户端为准:客户端与 CLI装完之后怎么算装好了
- 装完该看到什么 — 命令正常结束时退出码为 0,终端里没有 "command not found" 或 "permission denied"。
- 一行自检 — 按该仓库 README 里给出的最小用法跑一次;能按 README 的描述输出就是通了。
- 答不出来怎么办 — 这个项目没有写排错说明,去仓库的 Issues 搜 install。
安装命令取自该仓库;以上判据只描述可观察到的结果,不替代项目自己的文档。
- ★ 数取值
- 0 ★ · blink1217/trading-swarm-mcp · 取自 GitHub 搜索 API · (2026-11-19 20:00:48 核对)
- 来源明细
- 从该仓库的 GitHub 页面直接读取 · 条目类型:开源项目
- 收录与活跃
- 收录于 2026-11-18 · 最近见到 2026-11-20 · 共见到 3 次 · 被 1 个来源列出
- 评分构成
- 综合分 103 · 1 个来源 · repo
- 安装
- 安装命令见上方「安装与配置」。
常见问题(答案都来自我们的记录)
怎么安装?
该仓库给过一条安装命令,但我们还没核对出它属于这个条目;先别照抄,去它的 README 核对。
它适合谁 / 用来做什么?
我们把它归在这些标签下: MCP 服务。标签来自条目自身,不是我们强加的分类。
谁维护它、还在更新吗?
仓库
blink1217/trading-swarm-mcp;我们收录于 2026-11-18;最近见到 2026-11-20。这里没有"官方认证"或"质量评级"——我们只记录来源与时间。README 原文(英文,未翻译) · 在 GitHub 读完整 README →
# quant-swarm — Zero-Trust Backtest Forensics & Point-in-Time Market Data (MCP)
以下摘录自该仓库的 README(片段)
Audit trading backtests locally for lookahead bias, Newey-West + Benjamini-Hochberg overlap-adjusted significance, outlier concentration, and realistic transaction-cost decay—100% offline and free—and query as-filed Point-in-Time SEC EDGAR & macro signals with zero restatement leakage.
## 30-Second Quickstart (No Account or API Key Required Locally)
### Claude Code
bash
claude mcp add quant-swarm -- uvx quant-swarm
### Cursor / Claude Desktop / Windsurf (mcp.json)
json
"mcpServers": {
"quant-swarm": {
"command": "uvx",
"args": ["quant-swarm"]
### Remote Streamable-HTTP (claude.ai, ChatGPT, Web MCP Connectors)
text
https://mcp.1.21initiative.com/mcp
(Metadata & tools/list require no auth; tool execution uses OAuth 2.1 PKCE or Authorization: Bearer from 1.21initiative.com/mcp/.)
## First Prompt to Try
> "Run audit.backtest on ./returns.csv and ./trades.csv assuming 3 bps spread and 5 bps slippage across 25 parameter trials. Check whether net Sharpe survives 2x cost stress and Newey-West + Benjamini-Hochberg adjustment."
Sample Verdict Returned (0 credits, 0 bytes sent off your machine):
json
"tool": "audit.backtest",
"overall_assessment": "PROBLEMS_OBSERVED",
"metrics": {
"sharpe_gross_annualized": 2.14,
"sharpe_net_annualized": 0.31,
"newey_west_t_stat": 1.18,
"bh_q_value": 0.42,
"breakeven_cost_multiplier": 1.12,
"top5_pnl_share_pct": 68.4
## Default core Profile (12 High-Signal Tools, ~2.4k Schema Tokens)
uvx quant-swarm loads a lean 12-tool core profile so your coding agent never suffers tool-catalog context bloat:
| Surface | Tools Included in Default core Profile | Auth / Metering |
| Free Offline Backtest & Risk Forensics | audit.backtest (or audit.backtest_inline on HTTP), warden.audit_features, warden.cost_check, warden.validate_order, warden.prop_challenge_check, warden.defensive_posture, warden.explain_sizing, gym.probe_fragility | Free & 100% Offline (stdio needs zero token and makes zero network calls) |
| Hosted Point-in-Time (pit.) Data | pit.analysis, market.fundamentals_pit, market.insider_institutional_pit, market.regime | SWARM_MCP_ACCESS_TOKEN (free 250 calls/mo + 100 bonus Pro trial credits; one-time £19 / 10k or £149 / 100k credit packs, no subscription) |
### Specialist Profiles (--profile)
Need the full 48-tool specialist suites (step-by-step audit.preflight/quote/run/compare, portfolio optimization, symbolic alpha formula evaluation, or optional experimental Shadow Tournament benchmarking)?
bash
uvx quant-swarm --profile full # All 48 tools across Warden, Data & Gym
## 30-Second Quickstart (No Account or API Key Required Locally)
### Claude Code
bash
claude mcp add quant-swarm -- uvx quant-swarm
### Cursor / Claude Desktop / Windsurf (mcp.json)
json
"mcpServers": {
"quant-swarm": {
"command": "uvx",
"args": ["quant-swarm"]
### Remote Streamable-HTTP (claude.ai, ChatGPT, Web MCP Connectors)
text
https://mcp.1.21initiative.com/mcp
(Metadata & tools/list require no auth; tool execution uses OAuth 2.1 PKCE or Authorization: Bearer from 1.21initiative.com/mcp/.)
## First Prompt to Try
> "Run audit.backtest on ./returns.csv and ./trades.csv assuming 3 bps spread and 5 bps slippage across 25 parameter trials. Check whether net Sharpe survives 2x cost stress and Newey-West + Benjamini-Hochberg adjustment."
Sample Verdict Returned (0 credits, 0 bytes sent off your machine):
json
"tool": "audit.backtest",
"overall_assessment": "PROBLEMS_OBSERVED",
"metrics": {
"sharpe_gross_annualized": 2.14,
"sharpe_net_annualized": 0.31,
"newey_west_t_stat": 1.18,
"bh_q_value": 0.42,
"breakeven_cost_multiplier": 1.12,
"top5_pnl_share_pct": 68.4
## Default core Profile (12 High-Signal Tools, ~2.4k Schema Tokens)
uvx quant-swarm loads a lean 12-tool core profile so your coding agent never suffers tool-catalog context bloat:
| Surface | Tools Included in Default core Profile | Auth / Metering |
| Free Offline Backtest & Risk Forensics | audit.backtest (or audit.backtest_inline on HTTP), warden.audit_features, warden.cost_check, warden.validate_order, warden.prop_challenge_check, warden.defensive_posture, warden.explain_sizing, gym.probe_fragility | Free & 100% Offline (stdio needs zero token and makes zero network calls) |
| Hosted Point-in-Time (pit.) Data | pit.analysis, market.fundamentals_pit, market.insider_institutional_pit, market.regime | SWARM_MCP_ACCESS_TOKEN (free 250 calls/mo + 100 bonus Pro trial credits; one-time £19 / 10k or £149 / 100k credit packs, no subscription) |
### Specialist Profiles (--profile)
Need the full 48-tool specialist suites (step-by-step audit.preflight/quote/run/compare, portfolio optimization, symbolic alpha formula evaluation, or optional experimental Shadow Tournament benchmarking)?
bash
uvx quant-swarm --profile full # All 48 tools across Warden, Data & Gym
收录徽章(可选)
想把收录标记放进 README 就复制这行,可选,不影响收录。
[](https://anclum.com/zh/s/blink1217-trading-swarm-mcp/)
- 来源
- github · https://github.com/blink1217/trading-swarm-mcp
- 按领域浏览
- MCP 服务
- 收录于
- 今天
更多 MCP 服务器
所属领域与标签
每个标签都能点进去看同类的全部条目。数字是当前库里的真实条数。