llama-index-tools-algovault 0.1.0

AlgoVault ToolSpec for LlamaIndex — composite BUY/SELL/HOLD trade calls, market-regime classification, and cross-venue funding-arbitrage scans for crypto perpetual futures, backed by the AlgoVault MCP server.

LlamaIndex Tools: AlgoVault

Composite BUY / SELL / HOLD trade calls, market‑regime classification, and cross‑venue funding‑arbitrage scans for crypto perpetual futures — exposed as native LlamaIndex tools, backed by the AlgoVault MCP server.

One import, zero MCP wiring, keyless free tier by default. Built by AlgoVault Labs.

pip install llama-index-tools-algovault

Why

AlgoVault is an API‑first signal‑interpretation layer for AI agents: it turns raw market data into a single composite verdict with a queryable, public track record. This ToolSpec gives a LlamaIndex agent four AlgoVault tools out of the box — no API key required to start (free tier), and a one‑line api_key for paid tiers.

The performance the verdicts are scored against (PFE win‑rate) is published live at algovault.com/track-record and via the MCP signal-performance resource.

Quick start (branded ToolSpec)

from llama_index.tools.algovault import AlgoVaultToolSpec
from llama_index.core.agent.workflow import FunctionAgent
from llama_index.llms.openai import OpenAI

# Keyless free tier (100 calls / month). Pass api_key="..." or set
# ALGOVAULT_API_KEY for paid tiers.
algovault = AlgoVaultToolSpec()

agent = FunctionAgent(
    tools=algovault.to_tool_list(),
    llm=OpenAI(model="gpt-4o"),
    system_prompt="You are a crypto trading research assistant.",
)

resp = await agent.run("What's the current trade call for BTC?")
print(resp)

Call a tool directly:

algovault = AlgoVaultToolSpec()
print(algovault.get_trade_call("BTC"))                       # free
print(algovault.get_trade_call("ETH", timeframe="4h", exchange="BYBIT"))
print(algovault.get_market_regime("SOL", timeframe="1d"))    # paid tier
print(algovault.scan_funding_arb(min_spread_bps=5, limit=5)) # paid tier

Authentication

Mode	How	Tier
Keyless (default)	AlgoVaultToolSpec()	Free — 100 calls / month
API key	AlgoVaultToolSpec(api_key="...") or ALGOVAULT_API_KEY env	Paid tiers

The key (when set) is sent as an Authorization: Bearer <key> header to https://api.algovault.com/mcp.

Tools

Tool	Tier	Signature	Returns
get_trade_call	Free	coin, timeframe?, exchange?, include_reasoning=True, asset_class?	Composite BUY/SELL/HOLD verdict + confidence, price, indicators, regime, reasoning
get_trade_signal	Free	(same as get_trade_call)	Alias of get_trade_call, kept for back‑compat
get_market_regime	Paid	coin, timeframe="4h", exchange="HL"	Regime: TRENDING_UP / TRENDING_DOWN / RANGING / VOLATILE
scan_funding_arb	Paid	min_spread_bps=5, limit=10	Cross‑venue funding‑rate spreads (long one venue, short the other)

timeframe ∈ 1m 3m 5m 15m 30m 1h 2h 4h 8h 12h 1d (regime: 1h 4h 1d). exchange ∈ HL BINANCE BYBIT OKX BITGET ASTER EDGEX GATE MEXC KUCOIN PHEMEX BINGX HTX WEEX BITMART XT WHITEBIT. Defaults match the AlgoVault server (timeframe=15m, exchange=BINANCE for trade calls). On the free tier scan_funding_arb caps limit at 5.

Alternative: generic MCP client (no package)

AlgoVault is a standard Streamable‑HTTP MCP server, so any LlamaIndex agent can reach the same tools through the generic llama-index-tools-mcp adapter — by configuration alone, no AlgoVault‑specific code:

from llama_index.tools.mcp import BasicMCPClient, McpToolSpec

# keyless free tier
client = BasicMCPClient("https://api.algovault.com/mcp")
# paid tiers:
# client = BasicMCPClient(
#     "https://api.algovault.com/mcp",
#     headers={"Authorization": "Bearer <ALGOVAULT_API_KEY>"},
# )

spec = McpToolSpec(
    client=client,
    allowed_tools=["get_trade_call", "get_trade_signal", "get_market_regime", "scan_funding_arb"],
)
tools = spec.to_tool_list()          # or: await spec.to_tool_list_async()

This package (llama-index-tools-algovault) is the branded, zero‑config, schema'd wrapper of that same surface: one import, the four tools pre‑wired with typed signatures and docstrings, keyless by default. Use whichever fits — they call the identical MCP endpoint.

Listing

Per run‑llama's current contribution policy, new integrations are published to PyPI independently rather than added to the run-llama/llama_index monorepo (which no longer accepts new integration packages). This package is therefore distributed via PyPI and GitHub; a LlamaHub catalog entry is requested separately through the directory's "Request an Integration" flow.

Development

uv run -- pytest -m "not live"   # unit tests
uv run -- pytest -m live -s      # live keyless call against api.algovault.com/mcp
uv build                         # sdist + wheel

Links

• Website — https://algovault.com
• Track record (live PFE win‑rate) — https://algovault.com/track-record
• MCP server — https://api.algovault.com/mcp

---

Built by AlgoVault Labs. MIT licensed.