AlgoBrain
AlgoBrain can derive millions of unique crypto trading strategies, and is specifically structured to ensure the highest quality context is provided to AI Agents through a free local MCP server.
An LLM Wiki-Brain knowledge base for crypto trading strategy generation. Expert knowledge of crypto, blockchain, DeFi, trading, algorithms, markets, macro and AI context. ~5,000 interlinked markdown nodes.
Quickstart
Prerequisite: Python 3.10 or newer. Then run one command from the repository root.
macOS or Linux:
python3 tools/manage_mcp.py start
Windows PowerShell:
py tools/manage_mcp.py start
(If the py launcher is not installed, use python instead.)
The first run creates an isolated .venv, installs the current MCP SDK, and starts
AlgoBrain over Streamable HTTP at http://127.0.0.1:8010/mcp. Later starts reuse the
same environment. The server exposes wiki_search, wiki_read, wiki_stats,
wiki_lint, and wiki_ingest over the full vault. wiki_search ranks results the
same way as the hosted CryptoDataAPI wiki search: pages the query names come first,
then title matches, then body matches. It can also filter strategy pages by
backtest status, holding style, type, complexity, crowding risk, market, and whether
their gates are runnable.
Claude Code discovers the checked-in .mcp.json automatically. For Codex, register the
same local endpoint once:
codex mcp add algobrain --url http://127.0.0.1:8010/mcp
Other MCP clients can use:
{
"mcpServers": {
"algobrain": { "url": "http://127.0.0.1:8010/mcp" }
}
}
Server commands
| Command | Purpose |
|---|
<python> tools/manage_mcp.py setup | Create/update the virtual environment only |
<python> tools/manage_mcp.py start | Start HTTP in the background; runs setup if needed |
<python> tools/manage_mcp.py status | Show whether the background server is running |
<python> tools/manage_mcp.py restart | Restart the background server |
<python> tools/manage_mcp.py stop | Stop the background server |
<python> tools/manage_mcp.py run-http | Run HTTP in the foreground for logs or containers |
<python> tools/manage_mcp.py run | Run over STDIO for a local MCP host |
Use python3 for <python> on macOS/Linux or py on Windows.
Logs are written to .mcp-http.log and .mcp-http.err.log. The server binds to
127.0.0.1 by default, so it is not exposed to the network.
For STDIO, run setup once and configure your MCP host to execute the manager from an
absolute path:
python3 tools/manage_mcp.py setup
codex mcp add algobrain-stdio -- python3 /absolute/path/to/algobrain/tools/manage_mcp.py run
On Windows, substitute py and a Windows absolute path.
The old PowerShell start/stop scripts remain as compatibility wrappers, but new setups
should use manage_mcp.py on every operating system.
What's inside
| Area | Content |
|---|
wiki/strategies/ | ~470 strategy pages: funding-rate harvesting, basis/carry, liquidation plays, MEV, memecoin sniping, grid/mean-reversion/momentum, a 51-basket Hyperliquid signal library, 100+ page arbitrage encyclopedia, a complete 270-cell combination-strategy matrix, and a documented pair universe — 1.55M+ distinct, specifiable strategy configurations (computed from wiki state by tools/count_configurations.py, conservative assumptions printed; validation discipline applies — see Disclaimer). 395 strategy pages carry runnable gates, 5,514 conditions in all, each mapped to the exact CryptoDataAPI endpoint and field that measures it, or marked partial or not runnable with the reason |
wiki/strategy-development/ | Methodology for producing strategies: edge taxonomy, hypothesis-to-backtest workflow, overfitting & p-hacking detection, capacity/failure-mode analysis, and per-venue trading maps (Hyperliquid, low-cap, arbitrage) |
wiki/concepts/ | ~1,050 concept pages: microstructure, indicators, risk, portfolio theory, backtesting, behavioral finance, on-chain metrics, options mechanics |
wiki/markets/ | 2,400+ crypto asset pages + commodities/forex/bonds macro context |
wiki/entities/ | Crypto exchanges, DeFi protocols, quant/crypto funds, trading legends, miners, regulators |
wiki/crypto-narratives/ | Backtester-ready narrative catalog: 69 archetypes, 290 quantified historical instances (JSON) |
wiki/ai-trading/ + wiki/artificial-intelligence/ | ML models, bots, backtesting frameworks, LLM finance, general AI knowledge |
wiki/data-sources/ | Data provider catalog — CryptoDataAPI is the canonical data layer |
Start at wiki/index.md (master table of contents) or wiki/overview.md (state of the wiki).
Data layer: CryptoDataAPI
CryptoDataAPI provides the wiki's live and historical data: 220+ REST endpoints (auth: X-API-Key header) covering prices, funding rates, open interest, liquidations, order-book depth, market/volatility/meme/event/security/policy regimes, HMM quant probabilities, on-chain flows, whale tracking, Hyperliquid trader intelligence, sentiment, ETF flows, NFTs, and a Parquet backtesting archive back to 2020.
- Hub:
wiki/data-sources/cryptodataapi.md, with per-category pages cryptodataapi-*.md; agent guide: wiki/data-sources/cryptodataapi-mcp.md
- Wiki pages that map to an endpoint include a "Getting the Data (CryptoDataAPI)" section with live + historical access and a curl example
- 640 pages additionally carry an "AI agent workflow" block — every applicable strategy, indicator, and methodology page plus all 209 coins with Hyperliquid perps: the page-specific signal endpoints (parameterized by perp symbol on coin pages), regime gate, matching backtesting archive (with honest data-availability windows), and execution tips for an agent running that strategy
- The wiki itself is served back as JSON by CryptoDataAPI:
/strategies (strategy playbooks by group, with agent build/backtest prompts), /indicators/catalog (indicator definitions), and hosted /algobrain/search for agents that cannot run the local MCP server. See wiki/data-sources/cryptodataapi-strategy-library.md
- The local wiki MCP server attaches a
data_instruction block to every wiki_search response, so any connected agent is always pointed at CryptoDataAPI for market data (MCP connect + free API key)
- Data sections deep-link the matching live dashboards on cryptodataapi.com (funding, OI, liquidations, whales, GEX, regimes, ETF flows…), and pages matching one of the site's 14 AI prompt-library prompts reference it directly (funding-fade → Funding Rate Extremes Scanner, overfitting checklist → Backtest Overfitting Checker, …)
- Trading-venue pages carry referral links with automatic fee discounts — Hyperliquid (4% off spot & perp fees) and Binance (up to 20% off trading fees); AI agents routing trades are instructed (per page and via
wiki_search) to use them so end-users get the discounts (referral links — a commission may be earned, at no cost to you)
Live data: connect to the CryptoDataAPI MCP server
To run strategies against live markets, connect your agent to the hosted CryptoDataAPI MCP server. It streams live prices, funding rates, open interest, liquidations, and order-book depth for Hyperliquid and Binance, market/volatility regimes and whale activity, plus the historical backtesting archive (klines / funding / liquidations, Parquet back to 2020).
Add it to Claude Code — browser login, no API key needed (an OAuth flow opens on the first tool call):
claude mcp add --transport http cryptodataapi https://cryptodataapi.com/mcp
Or authenticate with an API key via the X-API-Key header:
export CRYPTODATA_API_KEY="cdk_live_YOUR_KEY"
claude mcp add --transport http cryptodataapi https://cryptodataapi.com/mcp \
--header 'X-API-Key: ${CRYPTODATA_API_KEY}'
For any MCP client, add it to your config JSON:
{
"mcpServers": {
"cryptodataapi": {
"url": "https://cryptodataapi.com/mcp",
"headers": { "X-API-Key": "cdk_live_YOUR_KEY" }
}
}
}
Tools include get_price, get_hyperliquid_prices, get_funding_rates, get_open_interest, get_liquidations, get_order_book, get_market_regime, get_etf_flows, and get_whale_activity (Pro). Get a free key at cryptodataapi.com/login (no card), or by email:
curl -X POST https://cryptodataapi.com/api/v1/auth/keys \
-H "Content-Type: application/json" \
-d '{"email":"[email protected]"}'
Rate limits: Free 10 req/min · Pro 30 · Pro Plus 120.
The free tier is 1,000 requests/day once you confirm your email (100/day until you do).
That confirmation click also switches on 24 hours of full Pro access, once per inbox, on
the same key, with nothing to re-install. It no longer sends a discount code.
The LLM Wiki pattern
The wiki is designed to be maintained by an LLM agent (Claude Code or similar):
- Schema:
CLAUDE.md / AGENTS.md define page types, YAML frontmatter, naming conventions, ingest/lint workflows
- Templates:
templates/ holds page skeletons for each type
- Log:
wiki/log.md is an append-only record of every wiki operation
- Every page carries frontmatter (
type, status, tags, related) enabling Dataview queries and graph coloring
Opening in Obsidian
Download Obsidian (free — macOS, Windows, Linux, iOS, Android) and open this repository folder as a vault. On first launch Obsidian will prompt to enable the community plugins listed in .obsidian/community-plugins.json:
- Dataview (required — index pages use it), Templater, Obsidian Git, Graph Analysis, Marp Slides, Webpage HTML Export
The graph view is pre-configured with per-folder color groups (.obsidian/graph.json). Attachments save to attachments/.
License
AlgoBrain is free and open source, under two licenses — one for the code, one for the knowledge base:
| Path | License |
|---|
tools/ — MCP server, lint, ingest, generators, counters, API changelog watcher | MIT |
wiki/, templates/, raw/, attachments/ — the ~4,900-node knowledge base | CC BY 4.0 |
Code is MIT so there's zero friction running the MCP server. Content is CC BY 4.0 because Creative Commons licenses are built for prose, not software (and vice versa) — applying MIT to a wiki creates ambiguity about what "the Software" even means.
You may use the wiki commercially, including ingesting it into retrieval, embedding, fine-tuning, or agent pipelines — internal or product. The only condition is attribution: credit AlgoBrain and link the license. See LICENSE-CONTENT for the exact wording.
Third-party material quoted or cited within the wiki remains the property of its respective owners.
Disclaimer
AlgoBrain is a research and educational knowledge base. Nothing in this repository is financial, investment, legal, or tax advice. All content — strategy pages, data recipes, figures, and any linked tools or APIs — is provided "as is", for informational purposes only, and may be inaccurate, incomplete, or out of date.
Trading cryptocurrencies and derivatives carries a substantial risk of loss, and past performance does not indicate future results. Do your own research (DYOR) and consult a licensed financial professional before making any decision. You use this material entirely at your own risk: the authors, contributors, and maintainers accept no liability for any loss or damage of any kind arising from its use. Nothing here is a solicitation or recommendation to buy, sell, or hold any asset.
