AlgoBrain is an expert-curated knowledge base of ~4,900 interlinked markdown nodes — strategies, concepts, assets and market narratives — served to AI agents over a local MCP server. It gives a language model the context to derive and analyse millions of crypto trading strategies, grounded on live CryptoDataAPI data.
A structured, interlinked wiki built for strategy generation — not a blog, not a docs site. Every node is designed to be read by a language model as high-signal context.
Funding-rate harvesting, arbitrage, liquidation plays, MEV exploitation, grid, mean-reversion and momentum — parameterized into 1.55M+ distinct specifiable configurations an agent can enumerate and adapt.
The building blocks: market microstructure, derivatives, on-chain mechanics, risk and edge taxonomy — the vocabulary a model needs to reason about why a strategy works.
Per-asset context across crypto, plus documented market narratives — 69 archetypes with 290 quantified historical instances so strategies are grounded in what actually happened.
Opens as a ready-to-explore vault with a color-coded knowledge graph and pre-configured plugins — browse the interlinks visually, or let an agent traverse them programmatically.
AlgoBrain runs a local Model Context Protocol server so any MCP-capable agent — Claude Code, Cursor and others — can search and read the wiki as native tools while it works.
Five tools are exposed over Streamable HTTP at http://127.0.0.1:8010/mcp:
The agent searches for relevant strategy and concept nodes, reads them for context, then pairs that knowledge with live and historical market data from CryptoDataAPI to specify, ground and backtest a strategy.
A knowledge base is only useful if its strategies can be tested against reality. CryptoDataAPI is AlgoBrain's canonical data layer — the wiki teaches the strategy, the API grounds it in real market data.
Live and historical prices, funding, open interest, liquidations, order-book depth, regime analysis and whale tracking — auth with an X-API-Key header.
That many wiki pages carry an AI agent workflow section with parameterized signal endpoints and a matching backtesting archive — read the strategy, then run it.
Clean Parquet history for regime-conditioned backtests, plus live dashboards on cryptodataapi.com linked throughout the knowledge base for verification.
Explore the data layer: API docs · backtest data · market regimes · whale activity. Pair AlgoBrain with the CryptoDataAPI MCP server so one agent has both the knowledge and the live data.
The whole vault follows an LLM-maintenance pattern: a schema defined in CLAUDE.md, node templates in templates/, and an append-only operation log — so an agent can grow the brain safely, node by node, without breaking the graph.
Written in Python (~93.5%) with PowerShell tooling (~6.5%), using MCP over Streamable HTTP and the Obsidian vault format.
Everything under tools/ is MIT-licensed. Use it, fork it, ship it.
The wiki/, templates/, raw/ and attachments/ are CC BY 4.0 — free to use with attribution.
Anyone building AI trading agents or researching crypto strategies with a language model. It gives the model a structured, expert-curated knowledge base to reason over instead of relying on whatever it happened to memorize.
The wiki itself reads standalone, but AlgoBrain is designed to be paired with CryptoDataAPI as its data layer — that's what turns a strategy idea into something you can specify, ground and backtest. A free API key is enough to start; deep history is Pro Plus.
It provides the knowledge and structure for an agent to generate them — ~445 strategy templates parameterized into 1.55M+ configurations, plus the concepts and market narratives to reason about which fit current conditions. The agent + CryptoDataAPI data does the generating and testing. We wrote about why that 1.5M number is a trap — the count is the easy part; the edge is not.
Clone the repo, install tools/requirements.txt, and run /start-servers (or the PowerShell script) to launch the MCP server on port 8010. Point your agent at it, and add the CryptoDataAPI MCP server alongside for live data.
An open-source knowledge base of strategies, concepts and market narratives — served over MCP, grounded on CryptoDataAPI data. Clone it and connect it to your agent today.
View on GitHub Add the data layer: MCP server → · Backtest data →