Engineering deep-dives, tutorials, and how-to guides for developers building with AI agents and MCP tools.
Most agent setups treat your notes as RAG fodder — chunk, embed, retrieve. TradeLoop's journal is different: the agent incrementally writes a markdown knowledge base of theses, trades, and lessons that compounds across sessions. Here's how it works, and why it's built on git + SQLite FTS5 instead of a vector store.
TradeLoop Engineering
Deep due diligence needs more than one search engine. TradeLoop's research gateway lets your agent use Perplexity, Firecrawl, and Grok behind a single Bearer token with shared keys and per-tier quotas — so the agent can dig into a name without you wiring up three accounts.
TradeLoop Engineering
Most sentiment tools are dashboards with a monthly bill. TradeLoop's version is the opposite: source skills for FinTwit (via Grok over X) and Reddit that your agent composes on demand. No standing pipeline, no merged feed — and that's deliberate.
TradeLoop Engineering
Ask Claude about a ticker and it answers from training data months stale. TradeLoop wires live Polygon, Finnhub, CoinGecko, FRED, and SEC EDGAR feeds into your agent over MCP, so it reasons over today's tape instead of last year's.
TradeLoop Engineering
The Model Context Protocol (MCP) is the open standard that lets AI assistants like Claude and ChatGPT call external tools — live quotes, indicators, earnings, SEC filings. Here's how it works, why it matters for traders, and how to use it today.
TradeLoop Team
Your market-data and broker API keys are the keys to your money and your edge. TradeLoop never stores them in plaintext — it uses a device-bound master key in the OS keychain plus atomic AES-256-GCM writes. Here's exactly how it works.
TradeLoop Engineering
What makes a good market-data layer for an AI agent? Live multi-asset feeds, precomputed indicators, a proactive monitoring loop, local-first key storage, and multi-client support. Here's what to look for — and how TradeLoop stacks up.
TradeLoop Team
A complete guide to giving Claude Code live market data — from how MCP works in Claude Code, to wiring up Polygon, Finnhub, indicators, and SEC filings, to fixing the most common setup problems.
TradeLoop Team
Nothing kills a research session faster than a stale OAuth token mid-task — your journal write fails, your broker read 401s. Here's how TradeLoop's background daemon detects expiring tokens and refreshes them before your agent ever sees an error.
TradeLoop Engineering
Should you hand-roll Polygon and Finnhub calls into your trading agent, or expose them over the Model Context Protocol? After building both, here's the technical and operational case for MCP — and where a custom integration still makes sense.
TradeLoop Engineering
Cursor supports MCP natively since version 0.45. Here's how to wire TradeLoop's market tools into Cursor so your AI pair can pull live quotes, compute indicators, check earnings, and read SEC filings without leaving your editor.
TradeLoop Team