quant-swarm
trading-swarm-mcp — the checkers that gate live capital, as MCP servers The pre-trade invariants and data assembly that guard live capital in the 1.21…
Что умеет
- Features.Build: Exact FEATURE_ORDER feature vector for symbol at session as_of, with provenance per field. Runs the no-lookahead guard (every feature must equal a fresh causal recomputation at as_of)
- Market.Pulse: Derived-only technical snapshot: trend, volatility, oscillator, volume, and regime signals. Never returns raw open/high/low/close/volume or provider quote values — only ratios, percentil
- Market.Sentiment: Derived-only sentiment/news snapshot for one symbol: headline count, earnings flag, and day-change bucket. Never returns raw quote values or headline text — quantized signals only. N
Какие данные видит
Нужен ли аккаунт
Нужен API-ключ из настроек сервиса
trading-swarm-mcp — the checkers that gate live capital, as MCP servers
The pre-trade invariants and data assembly that guard live capital in the 1.21 Initiative trading swarm, packaged as Model Context Protocol (MCP) servers you can run locally — in Cursor, Claude Desktop, or Claude Code — against your own strategies.
No provider credentials. No DRM. No order-placement code path. And no unlimited plans: capacity is bounded by what you pay for, on purpose.
What it does
Three servers, extracted from the same code that gates real money every market day:
| Server | Tools (Free) | Tools (Pro) |
|---|---|---|
| swarm-warden-mcp — pre-trade invariant & leakage auditing | validate_order, cost_check, explain_sizing, validate_genome, audit_features |
request_promotion_verdict |
| swarm-data-mcp — point-in-time data assembly | market_pulse, sentiment_pulse, regime_snapshot, cache_stats, offline_mode |
build_features, cache_warm, microstructure_snapshot, volume_forecast, screen_universe, cross_sectional_rank |
| swarm-gym-mcp — regime-fragility probing | — | label_regimes, probe_fragility, paired_preview, estimate_cloud_run |
- Warden — validate a proposed order against the house floors (25% position cap, 60% gross), audit a feature manifest for lookahead leakage, convert a gross edge into net-of-cost reality, explain the live C# risk engine's sizing math step by step, and validate a genome against the bounded schema.
- Data — derived-only market analytics (trend/volatility/oscillator/volume percentile ranks + causal regime labels), point-in-time feature vectors with provenance guards, order-flow imbalance / VPIN toxicity / effective-spread snapshots, volume forecasts expressed as ratios (never raw counts), and cross-sectional ranking.
- Gym — label causal regimes, replay a genome on a capped episode-seed matrix and surface per-regime fragility, and estimate the hosted tournament run. Promotions are never issued locally — undecidable outputs name the exact missing inputs and hand off server-side.
The data policy that makes this trustworthy
- Derived-only. Responses contain ratios, percentile ranks, labels, buckets, and counts. Raw open/high/low/close/volume, provider quote values, and headline text are never echoed back. Tools accept optional caller-supplied
barsfor symbols the shared cache doesn't carry. - Runs on your machine. Your orders, features, and genomes never leave it. No tool can place, cancel, or route an order — there is no order-placement code path.
- No provider credentials. Data flows through the hosted relay under your token; you never need an Alpaca or Finnhub key.
Plans & billing
| Free | Credit packs | |
|---|---|---|
| Cost | £0, instant token | £19 — 10,000 calls · £149 — 100,000 calls |
| Validity | monthly allowance (~250 calls) | 90 days from purchase, no auto-renewal |
| Symbols/call | 10 | 50 |
| Backfill | 365 days | full window |
| Tools | warden checkers + derived snapshots | everything, including the gym |
Pro is credit-based by design — you buy a fixed pool of calls, use them at your own pace, and buy again when you need more. There is no subscription and no unlimited plan; the credit pool on your token is the only capacity you can consume. Credits are metered per call, expire 90 days after purchase, and stack — buying again extends your window.
Quickstart
Get a free token at https://1.21initiative.com/mcp/ (instant), then add the servers to your client:
{
"mcpServers": {
"swarm-data": { "command": "uvx", "args": ["--from", "trading-swarm-mcp", "swarm-data-mcp"], "env": { "SWARM_MCP_ACCESS_TOKEN": "" } },
"swarm-warden": { "command": "uvx", "args": ["--from", "trading-swarm-mcp", "swarm-warden-mcp"], "env": { "SWARM_MCP_ACCESS_TOKEN": "" } },
"swarm-gym": { "command": "uvx", "args": ["--from", "trading-swarm-mcp", "swarm-gym-mcp"], "env": { "SWARM_MCP_ACCESS_TOKEN": "" } }
}
}
Как подключить
Как подключить quant-swarm к Claude
Адрес сервера, который нужно вставить в настройках Claude:
https://quant-swarm--blink-kt.run.tools
- Откройте Claude (сайт claude.ai или приложение для компьютера).
- Перейдите в Настройки → Коннекторы (в английском интерфейсе: Settings → Connectors).
- Нажмите Добавить пользовательский коннектор (Add custom connector).
- Вставьте адрес сервера, скопированный ниже, в поле URL удалённого MCP-сервера и нажмите Добавить.
- В чате нажмите + → Коннекторы и включите новый коннектор.
Пользовательские коннекторы доступны на тарифах Free, Pro, Max, Team и Enterprise. На Free можно добавить только один. В Team и Enterprise коннектор сначала добавляет владелец организации в Настройки организации → Коннекторы.
Авторизация
После нажатия Добавить откроется окно входа в сервис. Войдите под своим аккаунтом и подтвердите доступ. Пароль Claude не видит.
Как подключить quant-swarm к ChatGPT
Адрес сервера, который нужно вставить в настройках ChatGPT:
https://quant-swarm--blink-kt.run.tools
- Откройте ChatGPT в браузере (chatgpt.com). Нужен тариф Plus, Pro, Business, Enterprise или Edu.
- Один раз включите режим разработчика: Настройки → Приложения → Дополнительные настройки → Режим разработчика (Settings → Apps → Advanced settings → Developer mode).
- Откройте Настройки → Коннекторы и нажмите Создать (Create).
- Заполните форму: Название (любое), Описание (одна фраза, что делает сервис), URL MCP-сервера (скопируйте ниже).
- В поле Аутентификация выберите OAuth, если сервис требует вход, иначе Нет. Нажмите Создать.
- В новом чате откройте + → Приложения/Коннекторы и включите коннектор.
Название раздела у OpenAI менялось (Connectors → Apps/Plugins). Если пункт называется иначе, ищите «режим разработчика» в настройках. В рабочих пространствах Business/Enterprise администратор должен разрешить пользовательские коннекторы.
Авторизация
При первом использовании ChatGPT откроет окно входа в сервис. Войдите и подтвердите доступ.
Как подключить quant-swarm к Cursor
Адрес сервера, который нужно вставить в настройках Cursor:
https://quant-swarm--blink-kt.run.tools
Быстрый способ: нажмите кнопку Открыть в Cursor ниже, Cursor сам предложит добавить сервер.
Вручную:
- В Cursor откройте Settings → Cursor Settings → MCP и нажмите Add new global MCP server.
- Вставьте JSON, скопированный ниже, в открывшийся файл
~/.cursor/mcp.json(для одного проекта:.cursor/mcp.jsonв корне проекта). - Сохраните файл. Сервер появится в списке MCP; при необходимости пройдите авторизацию по кнопке рядом с ним.
Авторизация
Если сервис требует вход, рядом с сервером в списке MCP появится кнопка авторизации.
Список инструментов сервера (11)
Технические названия из tools/list. Нужны только разработчикам.
| features.build | Exact FEATURE_ORDER feature vector for symbol at session as_of, with provenance per field. Runs the no-lookahead guard (every feature must equal a fresh causal recomputation at as_of) and the point-in-time provenance guards. Tier-B/C fields without recorded point-in-time evidence are returned as UNSCORABLE — never neutral-filled with 0.0. |
| market.pulse | Derived-only technical snapshot: trend, volatility, oscillator, volume, and regime signals. Never returns raw open/high/low/close/volume or provider quote values — only ratios, percentile ranks (trailing 1y), and causal regime labels. Pass `symbols` to use the shared cache/relay, or pass your own `bars` rows (same shape as get_bars) for symbols we don't carry. This is the recommended tool for general market analysis; not investment advice. |
| market.sentiment | Derived-only sentiment/news snapshot for one symbol: headline count, earnings flag, and day-change bucket. Never returns raw quote values or headline text — quantized signals only. Not investment advice. |
| cache.warm | Backfill history for a universe (default: the swarm's 13-symbol universe). After warming, every finalized session is immutable in cache — replays cost zero API calls. Requires network and a valid SWARM_MCP_ACCESS_TOKEN; refused in cache.offline mode. |
| cache.stats | Cache inventory: rows per symbol, session bounds, depth in weeks, API-call accounting. |
| cache.offline | Toggle cache-only mode: no network access; reads served from cache, fetches refused. |
| market.regime | Free tool — causal regime label per symbol (high_vol / drawdown / chop / melt_up / gap_heavy), derived from point-in-time bars with no lookahead. |
| market.microstructure | Derived-only microstructure snapshot: order-flow imbalance, VPIN toxicity bucket, and effective half-spread — percentile-ranked within the series. Pro tool. |
| volume.forecast | Derived-only volume forecast expressed as a ratio to the trailing 20-session average, with a 10th-90th percentile interval. Never returns raw share counts. Pro tool. |
| market.screen | Ranked, derived-only screen over the symbol list — cross-sectional percentile ranks, never raw prices or volumes. Pro tool. |
| market.rank | Cross-sectional percentile rank of one derived metric across the symbol list — no raw metric values returned. Pro tool. |