Skip to content

TUTORIAL · MCP

Research an Indian stock with AI — without losing the evidence trail

This tutorial gives you one reusable workflow for researching an NSE/BSE company with an MCP-capable AI client. The examples work conceptually across Claude, ChatGPT and other clients that support remote MCP connections; the exact connection UI varies by client and plan.

The objective is not to get an AI stock tip. It is to make the assistant gather the same source-aware evidence you would otherwise collect manually — financials, peers, ownership, filings and freshness — before you form a view.

KEY TAKEAWAYS

  • Connect once to https://mcp.dalalos.in/mcp; then ask normal research questions instead of calling APIs by hand.
  • Start with identity/profile, then financial trend, peers, ownership and filings so later claims have a traceable base.
  • Ask for source and freshness on every material data block; DalalOS quote-style prices are end-of-day.
  • Keep the boundary explicit: no buy/sell calls, no target prices and no brokerage execution.

1. Connect the remote MCP

Open the DalalOS connection guideand choose your client. Claude supports remote MCP/connectors through its supported setup paths. ChatGPT supports MCP-backed custom apps/connectors on eligible plans and workspaces; its current availability and UI can differ by plan, so follow the live client-specific guide rather than copying an old screenshot.

claude_desktop_config.json
{
  "mcpServers": {
    "dalalos": {
      "type": "http",
      "url": "https://mcp.dalalos.in/mcp"
    }
  }
}

Canonical endpoint

Use https://mcp.dalalos.in/mcp. Do not paste an API key into a public prompt, tutorial or directory. Prefer the client’s supported OAuth flow when available; use a DalalOS bearer key only in clients that require the fallback.

2. Verify the connection with one grounded question

EXAMPLE PROMPTS

  • Show the latest available end-of-day quote for RELIANCE. Include the source and freshness timestamp, and tell me if the price is not intraday/live.

This first prompt checks three things at once: the AI can call DalalOS, the symbol resolves correctly, and the answer preserves the source/freshness contract. If it cannot show when the data was refreshed, do not move on to a longer research chain yet.

3. Run a four-part company research workflow

01

Company + financial trend

Resolve the company, then inspect reported revenue/profitability and supported mechanical ratios across consistent periods.

02

Peer frame

Find relevant peers and compare the same fields and dates rather than accepting an opaque ranking.

03

Ownership + pledge

Inspect promoter/FII/DII ownership and promoter pledge changes where available.

04

Filings + events

Use recent corporate disclosures and actions to investigate why a material number or ownership field changed.

4. Use prompts that ask for evidence, not verdicts

EXAMPLE PROMPTS

  • Show five years of revenue, operating profitability, ROE, ROCE and debt-to-equity for HDFCBANK using consistent periods. Include source/freshness context and flag missing fields instead of estimating them.
  • Compare TCS and Infosys on the same financial fields and periods. Do not rank them; show the underlying figures and state where definitions differ.
  • Show promoter, FII and DII holding changes plus promoter pledge history for this company. Then list recent filings that may explain material changes.
  • Summarise the latest available company disclosures as factual bullets. Separate disclosed facts from any inference, and link or identify the primary source where available.

5. Check the answer before using it

  • Freshness: is the price/reporting period recent enough for the question?
  • Comparability: are peer fields defined and dated consistently?
  • Missing data: did the model clearly say what was absent rather than filling a gap?
  • Primary evidence: for a material filing claim, open the original disclosure.
  • Inference boundary: distinguish a company-disclosed fact from the AI’s interpretation.

What DalalOS deliberately does not do

  • No live tick feed: quote-style market prices are end-of-day.
  • No brokerage order placement or portfolio trading actions.
  • No buy/sell calls, target prices or proprietary investment verdicts.
  • No promise that every company exposes every metric; disclosed/official-source coverage varies by dataset and issuer.

A better final prompt

Instead of “Should I buy this stock?”, ask: “Give me the strongest factual evidence for and against the business trend visible in the available data. Cite the underlying figures, sources and freshness; separate facts from inference; do not give a buy/sell recommendation.”

Once this single-company workflow is reliable, scale it into the research-desk or holdings-monitoring workflow.

FAQ

Common questions

Which AI clients can use DalalOS?

DalalOS exposes a remote MCP endpoint. Claude and other MCP-capable clients can connect using their supported remote-MCP flow. ChatGPT supports MCP-backed custom apps/connectors on eligible plans/workspaces; availability and UI vary, so use the current DalalOS connect guide for the client-specific path.

Does DalalOS provide live tick prices?

No. Quote-style DalalOS prices are end-of-day. The tutorial is for research workflows, not intraday trading or order execution.

Will the AI give buy or sell recommendations?

DalalOS itself provides no buy/sell calls, target prices or investment advice. Prompt the AI to show evidence, underlying figures, source and freshness rather than asking the data layer for a verdict.

How do I verify an AI answer?

Ask it to include source/freshness context, inspect the returned underlying figures, and open the original filing or source when a conclusion is material. Treat the AI as a research assistant, not the primary evidence.

Connect your AI to Indian stock market data

Sign in to DalalOS and connect your AI in one line of config. Free to start.

KEEP EXPLORING