I built SIMTEK MP1: an autonomous, risk gated options execution engine powered by @public API.
My engine reads live chains Greeks, scores setups, routes orders, reconciles broker state, and enforces risk gates before touching the account.
Seeing language I used privately show up elsewhere is a good reminder, access to a blueprint is not the same thing as owning the system.
The real work is in the receipts: code, logs, audits, failure cases, and the boring control layer that keeps autonomy from becoming chaos.
I published a short Substack on a lesson from my agent-governor’s first day in shadow mode:
Connected does not mean decision-grade.
Agents near high-risk systems should earn autonomy through evidence, not config flips.
jettmagnuson.substack.com/p/…
I’m exploring with Nvidia NemoClaw on simteks operating layer so the agent can inspect the system, read approved data, summarize guard health, explain the MCP decisions, and create follow up tasks on top the Kubernetes control-loop layer.
Copying the look of something to seem like you know what you’re doing only works on people who don’t know better yet - and they never stay that way for long. Flattered regardless.
The future is not “AI picking stocks.”
It’s agentic systems that can inspect live broker state, reason over risk, and execute only inside hard user-defined limits.
That’s exactly what I’m building with SIMTEK MP1.
AAPL trade, real money. @public
Discretionary me probably cuts this Monday.
SIMTEK MP1 didn’t.
The engine entered on a pullback, tracked the position through the drawdown, ran its EOD risk check, and held because the structure hadn’t broken.
Small P&L. Big proof of behavior.
I built SIMTEK MP1: an autonomous, risk gated options execution engine powered by @public API.
My engine reads live chains Greeks, scores setups, routes orders, reconciles broker state, and enforces risk gates before touching the account.
I built SIMTEK MP1: an autonomous, risk gated options execution engine powered by @public API.
My engine reads live chains Greeks, scores setups, routes orders, reconciles broker state, and enforces risk gates before touching the account.
The part i’m most proud are the LLMs can inspect the broker through read only MCP tools, they cannot place, modify, or cancel orders. trading authority stays isolated on the VPS engine.
This is not a strategy script.
It’s a live system: signals, options execution, portfolio reconciliation, analytics, alerts, and a public trade page showing real engine trades.
Portfolio:
darkskies-dev.vercel.app
I just open-sourced mcp-pulse — a drop-in observability SDK for MCP servers. One line of code: FastMCP -> ObserveMCP and every tool call gets tracked automatically.
GitHub: github.com/slimbiggins007/mc…
MCP servers are powerful, but most of them are still a black box. When a tool is slow or failing, authors often have no easy way to see what’s happening.
i built a mcp server that gives ai real financial analysis tools. not just just stock prices - rsi, macd, atr, fib, option chains, position sizing, and trade stats.
one command to install.
open source. works with claude, cursor, any mcp client.
there are 6 other finance mcp servers. they just all fetch stock prices.
fintools-mcp actually analyzes - it tells you the trend, sizes of your positions, calculates your r:r, and grades your trade history.