# VARRD — Complete Reference for AI Agents

> For the summary version, see llms.txt. This is the full technical reference.

## Quick Connect

```json
{ "mcpServers": { "varrd": { "url": "https://app.varrd.com/mcp" } } }
```

CLI: `pip install varrd && varrd edges`
Web: https://app.varrd.com
GitHub: https://github.com/augiemazza/varrd
Contact: management@varrd.com

## What VARRD Is

VARRD was built by a team from one of the most successful derivatives trading firms in Chicago history, alongside Princeton graduates and AI engineers. The firm's process for taking a raw trading idea and turning it into a validated, deployable system is what VARRD is built on — made accessible to anyone through AI.

Two products:
1. **Edge Library** — validated edges running 24/7 across futures, equities, and crypto. When an edge fires, you get everything: market, direction, entry, stop, target, hold period, formula, discovery story, and full audit trail.
2. **Research Engine** — test any trading idea in plain language. The AI generates hypotheses; a separate engine does all the math. Statistics are never fabricated.

## Questions You Can Ask

- "Is there an edge to going long Starbucks on their Red Cup Day?"
- "What happens to gold futures when yen and 10-year bonds make 20-day new highs on the same day?"
- "Take a look at crude oil on a 240-minute timeframe right now — diagnose how it's acting and find every time it's happened in the past."
- "Does it test out to short fast food companies during Lent?"
- "When corn and soybeans diverge by more than 2 standard deviations, is there a mean reversion trade?"
- "What happens to Bitcoin after a halving event?"

## MCP Tool Reference

### varrd_edges — Browse the Edge Library

The main product. Browse validated edges firing across futures, equities, and crypto.

**Parameters:**
- `depth` (integer, 0/1/2) — 0=free directory, 1=stats+trade levels, 2=full methodology
- `edge_id` (string) — specific edge ID for detail view
- `market` (string) — filter by market symbol (e.g. "ES", "GC", "AAPL")
- `status` (string) — "firing", "pending", "active"
- `direction` (string) — "LONG" or "SHORT"
- `timeframe` (string) �� "60min", "120min", "240min", "daily", "weekly"
- `asset_class` (string) — "futures", "equities", "crypto"
- `section` (string) — drill into: "setup_code", "horizons", "analytics", "occurrences", "view"

**Response format:** Plain text, designed for AI reading. Format varies by depth:
- depth=0: Market directory grouped by status (FIRING/PENDING/ACTIVE) with edge IDs
- depth=1: One-line per edge with direction, win rate, EV, stop/target, entry date, status tags
- depth=2 + edge_id: Full methodology card — TRADE, PERFORMANCE, INTEGRITY, DISCOVERY, FORMULA sections
- depth=2 + section: Section drill-down (table, analytics, code, signals list, or interactive chart URL)

**Pricing:**
- depth=0: FREE
- depth=1: $0.50 (all active edges)
- depth=2 + edge_id: $1 (single edge, permanent)
- depth=2 (all): $5 (every edge)
- Sections: free after depth=2 purchase

**Edge statuses:**
- FIRING: Signal confirmed. Enter at next bar's open. Actionable now.
- PENDING: Bar hasn't closed. Signal may fire when it does. Do NOT act.
- ACTIVE: Already in a trade from a previous signal.

### varrd_ai — Multi-Turn Research

Test any trading idea with real data. Multi-turn conversation — each response includes `context.next_actions`.

**Parameters:**
- `message` (string, required) — your idea, instruction, or follow-up
- `session_id` (string) — from previous call to continue conversation

**Typical flow (3-5 turns):**
1. "Test going long ES when RSI drops below 30" → charts the pattern
2. "test it" → runs statistical test
3. "show me the trade setup" → exact entry/stop/target

**Response includes:**
- `session_id` — pass back to continue
- `text` — AI response
- `widgets` — charts, test results, trade setups
- `context.has_edge` — true=edge found, false=no edge, null=still researching
- `context.next_actions` — follow these

**Cost:** ~$0.25/session

### autonomous_varrd_ai — Autonomous Discovery

Give it a topic, it discovers edges autonomously. Generates hypotheses from a market structure knowledge graph, tests them, returns results.

**Parameters:**
- `topic` (string, required) — e.g. "momentum on grains", "BTC weekend patterns"
- `markets` (array) — focus on specific markets
- `test_type` (string) — "event_study", "backtest", "both"
- `search_mode` (string) — "focused" or "explore"
- `asset_classes` (array) — "futures", "equities", "crypto"

**Cost:** ~$1/idea. Takes 1-3 minutes.

### search — Search Saved Strategies

Search YOUR saved strategies (not the edge library) by keyword or concept.

**Parameters:**
- `query` (string, required)
- `market` (string)
- `limit` (integer, default 10)

**Cost:** Free

### get_hypothesis — Strategy Detail

Full detail on a specific strategy you've created or saved.

**Parameters:**
- `hypothesis_id` (string, required)

**Returns:** formula, setup_code, direction, win_rate, ev_per_trade, sharpe, profit_factor, horizon_results, versions, performance analytics.

**Cost:** Free

### check_balance — Credit Balance

Check balance, see pricing, auto-detect completed payments.

**Parameters:** None

**Returns:** balance_cents, credit_packs, recovered_cents (if payment detected)

**Cost:** Free. Call after payment to confirm credits.

### buy_credits ��� Add Credits

Two payment paths: card (Stripe Checkout link) or crypto (USDC on Base).

**Parameters:**
- `amount_cents` (integer, default 500, min 500) — must be whole dollars
- `payment_method` (string) — "card" (default) or "crypto"
- `payment_intent_id` (string) — for crypto confirmation step

**Card flow:** Returns checkout_url → user clicks → pays → call check_balance
**Crypto flow:** Returns deposit address → send USDC → call buy_credits with payment_intent_id

**Cost:** Free to call

### get_briefed — Market News Briefing

Personalized news tied to your edge library. Requires edges in your library.

**Cost:** Credits

### reset_session — Kill Stuck Session

**Parameters:**
- `session_id` (string, required)

**Cost:** Free

## Edge Library Detail

Every validated edge includes:
- **Formula** — written in a domain-specific language, fully auditable
- **Discovery story** — what pattern was hypothesized, why it might work
- **Per-horizon results** — win rate, EV, p-value at every hold period tested
- **Post-discovery performance** — tracked separately from in-sample. Real out-of-sample.
- **Edge decay** — performance by time period, showing whether the edge is holding or fading
- **Monte Carlo** — simulated equity paths
- **Regime analysis** — performance in different VIX/trend regimes
- **Drawdown** — max, average, recovery factor
- **Return distribution** — skew, kurtosis, tail ratio
- **Interactive view** — browser link with chart, every signal, full stats (shareable, no account needed)

## What's Impossible on VARRD

- You cannot cherry-pick results — every test is counted
- You cannot peek at out-of-sample data — locked permanently after one use
- You cannot use future data — lookahead bias detected automatically
- You cannot fabricate statistics — the engine and AI are architecturally separated
- You cannot test the same thing twice — every test is fingerprinted
- You cannot disguise an old idea — embedding similarity detects overlap

## Data Coverage

- **Futures (CME):** Energies, metals, grains, currencies, indices, bonds — hourly through weekly, back to 1985
- **Equities:** Any US stock or ETF — hourly through weekly
- **Crypto (Binance):** BTC, ETH, SOL and more — hourly through weekly

## Credits and Payment

- Sign up at app.varrd.com → $2 free (email verification required)
- MCP/CLI agents start at $0 — call buy_credits for Stripe link ($5 min)
- Crypto: buy_credits(payment_method='crypto') — USDC on Base, fully autonomous
- Research: ~$0.25/session. Autonomous discovery: ~$1/idea.
- Edge library: free/$0.50/$1/$5 per query

## Getting Started

**Recommended path (get $2 free):**
1. Sign up at https://app.varrd.com (verify email)
2. Click avatar → Manage Agents
3. Copy the MCP config JSON
4. Paste into your MCP client config and restart

**Quick path (no signup):**
- Just connect with `{"mcpServers":{"varrd":{"url":"https://app.varrd.com/mcp"}}}`
- Call buy_credits when you need paid features

**Merge an existing agent:**
- If you started without an account, go to app.varrd.com → Manage Agents → Merge an Existing Agent → enter your passkey

## Guides

- [Best AI Trading Research Tools (2026)](https://www.varrd.com/guides/best-ai-trading-research-tools-2026.html)
- [MCP for Trading](https://www.varrd.com/guides/mcp-trading.html)
- [How to Trade with Claude Desktop](https://www.varrd.com/guides/claude-desktop-trading.html)
- [AI-Powered Backtesting](https://www.varrd.com/guides/ai-backtesting.html)
- [Event Studies in Trading](https://www.varrd.com/guides/event-study-trading.html)
- [Overfitting Prevention](https://www.varrd.com/guides/overfitting-prevention.html)
- [Out-of-Sample Testing](https://www.varrd.com/guides/out-of-sample-testing.html)
- [Stop-Loss Optimization](https://www.varrd.com/guides/stop-loss-optimization.html)
- [What Is an Edge in Trading?](https://www.varrd.com/guides/what-is-an-edge-in-trading.html)
- [How to Test a Trading Strategy](https://www.varrd.com/guides/how-to-test-a-trading-strategy.html)
- [Crypto Trading Edges](https://www.varrd.com/guides/crypto-trading-edges.html)
- [Quant AI and MCP](https://www.varrd.com/guides/quant-ai-mcp.html)