Technical notes
What we learn while building AI agents, MCP servers and the tools around them.
Vous cherchez nos services ? Retour à l’accueil en français.
-
Track whale positions and liquidation clusters on Hyperliquid
What whale and liquidation data Hyperliquid exposes, how to read where forced selling stacks up, and how to turn it into alerts an agent can watch in real time.
-
Hyperliquid funding rate API: funding, OI, predictions
Pull current, historical and predicted funding plus open interest from the Hyperliquid API, with the gotchas on sign and sampling that cost us a day.
-
Do more MCP tools make your agent worse?
Tool schemas eat context and bias selection. Lessons from a 47-tool MCP server: naming, disambiguation, grouping, and when to split into two servers.
-
MCP transports: local stdio vs remote HTTP, how to choose
Should an MCP server ship as a local stdio binary or a hosted HTTP endpoint? A decision matrix on data locality, auth, updates and install friction.
-
How to give your AI agent persistent memory across sessions
Stateless agents forget everything between runs. The three ways to add persistent memory, and how to wire a local MCP memory server into Claude or Cursor.
-
Local-first memory beats a vector bill
Cloud RAG charges you rent on your own thoughts. A local index in SQLite and LanceDB, with embeddings from Ollama, is faster, private, and yours.
-
What is MCP? A complete guide to Model Context Protocol
MCP is the open standard that lets any AI agent call external tools and data. How it works, why it matters, and what is worth building with it.
-
Polymarket + Hyperliquid through one MCP: a practical guide
Expose live Polymarket markets, Hyperliquid funding, open interest and whale flow to Claude, Cursor or any tool-using LLM through one MCP server.
-
Cross-venue divergence: the alpha that nobody runs manually
Three venues price the same event differently. The disagreement is the signal. Here's how PredMCP makes it tractable for an agent.