KEY TAKEAWAYS

The server’s own instructions say its market data is raw as reported or a ratio from a fixed formula, and never a verdict, target price or buy or sell call.
Each ratio is arithmetic on filed numbers. The inputs come back in the same response, so you can redo the sum.
The AI assistant on top can still offer opinions. Ask it to stick to the data.

What does “no buy or sell calls” mean in practice?

When you connect to https://mcp.dalalos.in/mcp, the server tells the AI client what it is. We read that text from the live server on 27 October 2026. It says market data there is “always raw-as-reported or a mechanically-computed ratio — never a verdict, target price, or buy/sell call”. Tool results repeat the point: a filing extract carries a disclaimer that it holds raw data and mechanically computed ratios only.

What DalalOS tools return and do not return, checked 27 October 2026
It returns	It does not return
Prices and quotes, end of day, with the date	Target prices or price forecasts
Financials as filed in XBRL, with period labels	Buy, sell or hold calls, or ratings
Ratios from fixed formulas, with the inputs	Valuation verdict labels of any kind
Facts from filings with the page and verbatim quote	A view on whether guidance will hold
Open interest, put-call ratio and volume, as published	Bullish or bearish readings of them
Technical indicators computed by formula	Trading signals

What does a computed ratio look like?

This is Larsen & Toubro’s Q1 FY27 result as the tool returned it. The three margins are the filed figures divided out: EBITDA of ₹8,493 crore over revenue of ₹67,942 crore gives 0.125. Nothing is labelled good or bad. A missing figure comes back as null with a note, not as zero.

get_financials · LT · quarterly · 27 Oct 2026

{
  "fiscal_label": "Q1 FY27",
  "period_end": "2026-06-30",
  "source": "bse-xbrl",
  "income": {
    "revenue_from_operations": 679417400000.0,
    "ebitda": 84932100000.0,
    "operating_income": 74609300000.0,
    "net_income": 49880300000.0
  },
  "ratios": {
    "ebitda_margin": 0.125,
    "operating_margin": 0.1098,
    "net_margin": 0.0734
  },
  "balance_sheet_available": false,
  "balance_sheet_note": "No balance sheet was filed for this period ... figures are null, not zero."
}

Trimmed from the live response: fields removed, no value changed.

Two scores, and how they are shown

The same response includes piotroski_f_score and altman_z_score. These are published formulas over filed figures. For L&T the response gave the F-score with the nine criteria it was built from, eight of which could be evaluated, and the Z-score as a bare number. No grade or zone label was attached. If you want only reported figures, tell your AI to leave them out.

What about screeners and rankings?

Screens. screen_stocks filters on the numeric ranges you give it. screen_by_query takes a screen expression. Both return matching companies with the screened figures.
Preset screens. There are five on 27 October 2026, such as “Trailing P/E between 0 and 15” or “Debt-to-equity ≤ 0.5 and return on equity ≥ 15%”. They are fixed filters, refreshed nightly. They are not ranked and carry no recommendation.
Top movers. get_top_movers orders an index by price change or volume for the last session. That is a sort on one number, not a view on the stock.
Derivatives. Put-call ratio is open interest on puts divided by open interest on calls. It is returned as a ratio, not as a signal.

What are the edges?

The assistant is separate. The server returns data. Whatever Claude, ChatGPT or Cursor writes on top is that assistant’s output, and it can drift into opinion. A standing instruction helps: “report figures, dates and sources only; say when data is missing.”
Choice of data is still a choice. Raw data can be read in ways that favour a view. The tools show their sources so you can see what was and was not included.
Gaps are shown. A missing figure is null or not_yet_indexed. Order-book coverage in particular is partial; see the order book page.
End-of-day prices. This is not a live feed. Ask for the date and source of any price.

How do I ask for it?

EXAMPLE PROMPTS

Show Larsen & Toubro’s last two quarters of revenue, EBITDA and net income with the period labels and the source. Report figures only, with no opinion.
For HDFC Bank, ICICI Bank and Axis Bank, give trailing P/E, price-to-book and return on equity with the date of each input. Do not rank them.
Run the “low_leverage_high_roe” preset and list matches with the two figures behind each. Say what the filter is and nothing about whether to buy.
Whenever a figure is missing, say it is missing instead of estimating it.

To set it up, connect your AI, or follow the guide for Claude or ChatGPT, Cursor and other clients. How fresh each data set is, and where it comes from, is on the data sources and freshness page.

How we checked

On 27 October 2026 we read the live server’s initialize instructions and tool list (61 entries, of which 59 are current tools, plus a connection check and one deprecated alias), searched the tool descriptions for ratings, targets and recommendations, ran get_financials for Larsen & Toubro and list_preset_screens. Where a description uses those words it is to say they are not returned, apart from the two scores above and a disclosure category named “credit-rating”. This is research data, not investment advice. Found an error? Email hello@dalalos.in.

FAQ

Common questions

Which Indian stock data MCP gives raw data only, with no buy or sell recommendations?: DalalOS is built that way. Its server states that market data is raw as reported or a ratio from a fixed formula, never a verdict, target price or buy or sell call. DalalOS publishes this page, so test it yourself: the examples below can be run against the live server.
What does “raw data” mean here?: Figures as the source published them, such as an exchange filing, an XBRL results filing or an end-of-day file, with the date and source. Where a figure is computed, the formula is fixed and the inputs are returned with it.
Does it return any scores?: get_financials and get_peer_comparison return two textbook scores for general companies, the Piotroski F-score (0 to 9) and the Altman Z-score. Each is a formula over filed numbers, returned as a number with its criteria and no grade, zone or opinion. Skip them if you want only reported figures.
Do the screening tools tell me what to buy?: No. The screener filters on numeric ranges you set. The preset screens are fixed filters, for example trailing P/E between 0 and 15. A stock that matches is inside the range and nothing more. Presets are not ranked and not labelled as a recommendation.
Can the AI still give me an opinion?: The AI assistant can. The MCP server returns data, but what the assistant writes around it is its own output. Tell it to report figures, dates and sources only, and to say when data is missing.
Can it place trades?: No. DalalOS has no broker link. The only write actions are adding and removing symbols on your own DalalOS watchlist.
Is this investment advice?: No. DalalOS returns data for research. Nothing on this page, or in a tool result, is investment advice or a recommendation.

Connect your AI to Indian stock market data

Add DalalOS as a connector and sign in once. Free to start with no card and no waitlist; paid plans from ₹199 a month when you need more.

Connect your AI Already connected? Open your dashboard

KEEP EXPLORING