What Is an MCP Server for Stocks? How It Works for Indian Markets

Tapetide is an MCP server for Indian stocks: it lets an AI assistant like Claude or ChatGPT call tools that return real NSE and BSE numbers, so it answers from data instead of memory. An MCP server for stocks is the piece that sits between the assistant and a market database. It publishes a list of tools, the assistant picks the ones a question needs, and the results come back into the chat. Tapetide's server lives at https://mcp.tapetide.com/mcp and covers about 8,000 listed companies, including small caps and SME listings, plus NSE indices like Nifty 50 and Bank Nifty. This explainer covers what MCP is, how a single tool call works, why it matters for Indian stock research, a worked example, and what to check before you trust an answer.

#What MCP (Model Context Protocol) is

MCP is an open protocol that lets AI assistants call external tools in a standard way. A server describes its tools, each with a name, a purpose and typed inputs. The client (Claude, ChatGPT, Cursor and others) shows those tools to the model. When a question needs data, the model picks a tool, the client runs it, and the result goes back into the conversation. One server works across every client that speaks the protocol, so you set it up once per app rather than writing an integration for each.

#Why an MCP server matters for Indian stock research

Language models know a lot about Indian companies in general. They are poor at specific, changing numbers. Ask for a mid cap's latest quarterly profit or this month's FII flows and you get one of three outcomes: a refusal, a stale figure from training data, or a confident number that is simply wrong.

A stock market MCP server fixes the input side. The AI fetches the figure, then reasons about it. That changes what you can trust it with:

• Comparisons across many companies. The AI pulls the same metrics for ten stocks without you opening ten tabs.
• Small caps and SME stocks. These are thin in training data. With tools, coverage matches large caps.
• Time-sensitive context. Shareholding changes, pledges, bulk deals and flows are fetched at question time.
• Traceable answers. You can ask for the period and date behind every number and check it on the matching stock page, such as /stocks/RELIANCE.

#How a tool call works, step by step

It helps to see what happens between your question and the answer. Take "What is HDFC Bank's promoter holding?"

1. The client lists the tools. When you enable the server, the client (Claude, ChatGPT, Cursor) asks it for its tool list. Each tool comes with a name, a one-line purpose and a typed input schema, such as a stock symbol and a number of quarters.
2. The model chooses. The model reads your question and the tool descriptions, and decides it needs search_stocks to resolve "HDFC Bank" to a listing, then get_shareholding for the holding pattern.
3. The client runs the call. The client sends the request to the server with your sign-in attached. Depending on your settings, it may ask you to approve the tool first.
4. The server answers with data. The server returns structured results: promoter, FII, DII and public holding by quarter, with the quarter end date.
5. The model writes the answer. The numbers go back into the conversation, and the model writes its reply from them. If it needs more, such as a pledge check, it repeats the loop with another tool.

Two things follow from this design. The model never sees the database, only the tool results, so the answer is as current as the data the server returns. And every number has a source call, so you can ask the assistant which tool and which date each figure came from.

#What kinds of tools a stock MCP server exposes

Servers differ, but a research server for Indian stocks usually covers a few families. Tapetide's fall into seven:

• Find and screen: resolve names, symbols, BSE codes and ISINs, and filter the market on fundamental ratios or technical signals.
• Company research: profiles, quarterly and annual financials, shareholding, pledges, forecasts, earnings call summaries and filings.
• Prices and indices: latest quotes, daily price history, index performance and India VIX.
• Market and flows: FII/DII activity, sector-wise foreign flows, market breadth and valuations. The same flows are charted on the FII/DII data page.
• Options: end-of-day index option chains, implied volatility history and derived figures such as put-call ratio.
• Scores and point-in-time: the composite stock score explained in the score methodology, plus "as of" tools for index membership and historical identifiers that keep backtests free of survivorship bias.
• Portfolio: read and edit your Tapetide watchlist and research portfolio.

For every tool name in each family, with prompts that exercise them, see the NSE and BSE MCP server tools guide.

#A worked example conversation

Here is how a typical exchange plays out in Claude with Tapetide enabled.

You: Find profitable mid cap pharma companies with low debt, then compare the top three on growth and promoter holding. Include dates.

What the AI does:

1. Calls screen_stocks with sector pharma, a mid cap market cap range, positive net profit and debt to equity below a threshold.
2. Calls get_financials for the top three results to get revenue and profit growth over recent quarters.
3. Calls get_shareholding and get_promoter_pledge for the same three.
4. Calls get_batch_quotes for current prices and valuation context.

What you get back: a table with each company's revenue growth, profit growth, operating margin, promoter holding trend and pledge status, with the quarter end date for each fundamental figure and the observation date for prices. The AI then summarises differences, for example one company growing faster but with falling promoter holding.

That took about six tool calls. On the free plan you can run several such questions a day.

#What to check before you trust an answer

Tools make answers better grounded, not automatically right. A few habits catch most problems.

• Ask for dates. Add "include the period or observation date for each figure" to your prompt. A quarterly number and a price from today are both correct but not comparable without dates.
• Ask which tool it used. If a figure looks off, ask the assistant which tool returned it. If it cannot name one, the number came from its training data.
• Spot-check one number. Open the matching stock page, such as /stocks/TITAN, and compare one figure. If that matches, the rest of the table usually does too.
• Watch for silent gaps. A very recent listing may have little history, and a screen can return fewer results than you expect. Ask how many companies matched before it ranked them.
• Keep research and trading apart. A research server tells you what the numbers say. It does not know your position size, tax situation or timing, and it does not place orders.

For a ready list of prompts to try once you are connected, see the example prompts in the NSE and BSE MCP server guide.

#How to connect

The endpoint is https://mcp.tapetide.com/mcp over Streamable HTTP with Google sign-in.

• Claude.ai: Settings, Connectors, Add custom connector, name Tapetide, paste the URL, sign in with Google, enable in a chat. Set Always allow under Tool permissions to skip per-tool prompts. Guide: /mcp/claude-ai.
• ChatGPT: Settings, Apps, turn on Developer mode, create an app with the URL and OAuth, sign in, select Tapetide in chat. A workspace admin may need to allow it. Guide: /mcp/chatgpt.
• Cursor, VS Code, Claude Code: add the server to your config.

{
  "mcpServers": {
    "tapetide": {
      "url": "https://mcp.tapetide.com/mcp"
    }
  }
}

In Claude Code: claude mcp add --transport http tapetide https://mcp.tapetide.com/mcp. Guides: /mcp/cursor, /mcp/vscode, /mcp/claude-code, /mcp/codex.

For stdio-only clients such as Claude Desktop, create a token at /settings/tokens and run bunx tapetide-mcp with TAPETIDE_TOKEN set (/mcp/claude-desktop). Every client is listed on the Indian stock market MCP server page.

#Who it is for

Retail investors get a research assistant that can pull fundamentals, ownership and flows for any listed company in plain language. Good for shortlisting, comparing peers and checking a tip against actual numbers before you spend time on it.

Developers get a typed tool surface for Indian equities that works inside coding agents. Use it to prototype screens, explore data before writing a backtest, or build agent workflows without maintaining your own data pipeline. Point-in-time tools help keep historical tests honest.

#Free plan and freshness

The free plan includes 50 successful tool calls per day, with no monthly cap. Higher limits are on the pricing page.

The server is not a live trading feed. Quotes are the latest available prices, financials and shareholding follow reporting periods, and index option chains are end-of-day snapshots. Ask the AI to include the observation date with every number. Data is aggregated from multiple licensed and public sources and structured by Tapetide.

Everything the AI produces from these tools is research, not a recommendation to buy or sell.