quantustik

Automated research routine on Claude

One prompt, run on a schedule, so every session starts from the same read.

Quantustik is a risk-first market-intelligence layer for humans and AI agents, delivering calibrated probabilistic forecasts, transparent track records, and machine-readable decision support through REST and MCP. A routine is that layer pulled on a fixed cadence instead of one question at a time: the same market conditions read, the same forecast check on the tickers you care about, the same insider-filing scan, every time, without you remembering to ask.

Disclaimer. Everything below is educational: how to point an AI client at Quantustik's own data and read the result. Nothing here is investment advice, a solicitation to buy or sell any security, or a personalized recommendation. Signals and forecasts are probabilistic model outputs with explicit uncertainty, not promises. Write your own prompts to ask for data, analysis, and explanation, never "tell me what to buy." Full text: Disclaimer & Terms of Use.

1. Connect once

If you have not already, add the Quantustik MCP connector (same setup as the standalone assistant): Settings → Connectors → Add custom connector → https://quantustik.com/mcp. Full walkthrough: Build your own market assistant on Claude.

2. Write the routine as one prompt

Naming the tools and their order keeps the routine reproducible: the same steps run the same way every time, instead of Claude improvising a different check each morning.

Run my morning market check:
1. Call get_market_regime and summarize the buying-conditions score and verdict in one sentence.
2. For AAPL, MSFT, and NVDA, call get_forecast with horizon '3mo' and report the probability of an up move plus the 90% confidence band for each.
3. For the same three tickers, call get_insider_pulse and flag any showing cluster buying.
4. Finish with one plain-English paragraph tying it together. Do not tell me what to buy; report what the data and the model say, and note the uncertainty.
Illustrative example. A run might read back: “Market conditions: score NN, Neutral. AAPL 3-month: NN% chance of an up move, band −X% to +Y%. MSFT 3-month: NN%, band −X% to +Y%. NVDA 3-month: NN%, band −X% to +Y%. No cluster insider buying flagged on any of the three. Overall: mixed, no strong directional read this week.” The values above are placeholders; the actual figures change daily, so run the routine yourself for the current reading.

3. Put it on a schedule

Where your Claude plan includes recurring or scheduled tasks, point one at the prompt above and give it a fixed time (for example, weekday mornings before the open). Where it does not, the prompt still works pasted by hand; it is the same routine, just started manually.

Congress trades are not an MCP tool yet

Congressional trade disclosures are real, public STOCK Act filings, and Quantustik serves them over REST (https://quantustik.com/api/v1/congress/feed), but they are not yet exposed as an MCP tool while the underlying feed is upgraded to source directly from House/Senate filings. If your Claude client can fetch a URL, ask it to summarize that feed directly; otherwise check quantustik.com/congress yourself alongside the routine above.

More ways to run this

Disclaimer. Everything below is educational: how to point an AI client at Quantustik's own data and read the result. Nothing here is investment advice, a solicitation to buy or sell any security, or a personalized recommendation. Signals and forecasts are probabilistic model outputs with explicit uncertainty, not promises. Write your own prompts to ask for data, analysis, and explanation, never "tell me what to buy." Full text: Disclaimer & Terms of Use.