Cursor MCP Stock Data: Indian Stocks in Cursor, VS Code, Codex and Claude Code

If you write code in Cursor, VS Code, Codex or Claude Code, you can give your agent direct access to Indian stock data through MCP. Tapetide MCP is a remote server that exposes more than 50 tools covering about 8,200 NSE and BSE listed companies, plus indices such as Nifty 50 and Bank Nifty. Once connected, the agent can pull financials, price history and shareholding while it writes your scripts, notebooks and reports. This guide covers the config for each editor, auth options, rate limits, and three workflows for building stock analysis with Cursor MCP or any other agent.

#What you get as a developer

The server runs at https://mcp.tapetide.com/mcp over Streamable HTTP. Your agent sees a list of tools with typed input schemas and decides which to call. The ones developers use most:

• search_stocks to resolve a company name, symbol, BSE code or ISIN
• get_financials for quarterly and annual statements
• get_price_history for daily OHLCV series
• get_batch_quotes and get_stock_quote for latest available prices
• screen_stocks (326 fundamental ratios) and screen_stocks_technical (RSI, MACD, moving averages, volume)
• get_shareholding, get_fii_dii_flows, get_corporate_actions
• get_watchlist and get_user_portfolio for your saved Tapetide research lists

Data is aggregated from multiple licensed and public sources and structured by Tapetide. It is not a live trading feed. Quotes are the latest available prices, financials follow reporting periods, and index option chains are end-of-day snapshots.

#Add Tapetide MCP to Cursor

Create .cursor/mcp.json in your project (or use the global MCP settings) with the remote URL form:

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

Cursor opens a Google sign-in the first time and stores the OAuth session. If you would rather use a personal token, for example on a shared machine or in a headless setup, send it as a header:

{
  "mcpServers": {
    "tapetide": {
      "url": "https://mcp.tapetide.com/mcp",
      "headers": {
        "Authorization": "Bearer YOUR_TOKEN"
      }
    }
  }
}

Generate the token at /settings/tokens. Keep it out of version control: add the config file to .gitignore, or keep the token-based entry in your global config. Full steps are on the Cursor setup page.

#VS Code MCP for Indian stocks (Copilot agent mode)

VS Code reads MCP servers from a workspace or user config file. Add the same remote URL, start the server from the config file or the command palette, then switch Copilot Chat to agent mode. The Tapetide tools show up in the tool picker, where you can enable or disable individual tools per chat. VS Code handles OAuth for remote servers, and the Bearer header form works too. The VS Code setup page has the exact file format for the current release.

#Codex and Claude Code

Both are terminal agents that work directly in your repo.

For Claude Code, one command registers the server:

claude mcp add --transport http tapetide https://mcp.tapetide.com/mcp

Run /mcp inside Claude Code to sign in and confirm the tools are listed. See /mcp/claude-code for project vs user scope.

Codex registers remote MCP servers in its config file with the same URL. The Codex setup page shows the current syntax and how to pass a token if you skip OAuth. Once connected, Codex MCP finance workflows look the same as in any other agent: you describe the output, the agent picks the tools.

#Workflow 1: a financials comparison table in markdown

This is a good first test because it touches search, data and file writing in one step.

Use the tapetide MCP tools. Resolve TCS, Infosys, HCL Technologies, Wipro and
Tech Mahindra with search_stocks, then call get_financials for each one.
Write docs/it-services-comparison.md with a table of the last 4 quarters:
revenue, operating profit, net profit and operating margin. Include the
reporting period for every number and note the date you fetched the data.

Expect around six tool calls: one or more searches plus five financials lookups. Review the table against the company pages for a couple of numbers before you trust the pattern.

#Workflow 2: a notebook from price history

Agents are good at turning a data pull into a reusable notebook.

Call get_price_history for HDFCBANK and ICICIBANK for the last 2 years of
daily data. Save the results as CSV under data/, then create
notebooks/bank-returns.ipynb that loads the CSVs with pandas, plots
normalised closing prices, and computes 20-day rolling volatility.
Print the last date in each series at the top of the notebook.

The agent fetches through MCP once and writes the data to disk. The notebook then runs offline against the CSV, so re-running cells costs no tool calls.

#Workflow 3: a watchlist check script

Read my Tapetide watchlist with get_watchlist. For each stock, get the
latest quote with get_batch_quotes and the latest shareholding with
get_shareholding. Write a markdown report to reports/watchlist.md that
flags any stock where promoter holding fell quarter on quarter or where
the price is more than 10% below its 52-week high. Include observation dates.

A few more prompts that work well in an editor:

Use screen_stocks to find NSE stocks with ROE above 20% and debt to equity
below 0.5, then write the results to a CSV with symbol, sector and both ratios.

Pull get_fii_dii_flows for the last 30 sessions and generate a matplotlib
chart script under scripts/fii_dii.py. Hardcode nothing; save the data as JSON first.

Using screen_stocks_technical, list Nifty 50 stocks with RSI below 30 and
explain in a comment block which fields the tool returned.

Portfolio and watchlist tools read and edit your Tapetide research records only. Nothing places orders at a broker. Everything the agent produces is research output, not a recommendation to buy or sell.

#OAuth vs token auth

For stdio-only clients there is also a local helper: run bunx tapetide-mcp with TAPETIDE_TOKEN set in the environment. Most coding agents support remote servers, so you rarely need it.

#Rate limits on the free plan

The free plan gives 50 successful tool calls per day and up to 1,000 per month. Coding agents can burn through that faster than chat users because they loop. A prompt that compares 20 companies may use 20 or more calls.

Ways to stay within budget:

• Prefer get_batch_quotes over many single get_stock_quote calls
• Ask the agent to save fetched data to disk and reuse it
• Narrow screens before asking for per-stock detail
• Tell the agent the call budget in the prompt ("use at most 8 tool calls")

If you need more headroom, see pricing.

#How MCP differs from calling a REST API

With a REST API you read docs, pick endpoints, write a client, handle auth and parse responses. With MCP the agent does that discovery itself:

• No SDK. The server publishes tool names and JSON schemas. The agent reads them at connect time.
• The agent chooses tools. You describe the outcome ("compare margins for these five"), and the agent decides to call search_stocks and then get_financials.
• Results land in context. The agent can reason over the data, ask a follow-up call, then write code or files.
• Same auth, same data. Freshness is identical to the site; MCP is another way in.

The trade-off is predictability. An agent may take a different path each run. For a production job that must run the same way every night, write the final pipeline against fixed requests. MCP shines in the exploratory phase: figuring out which data you need, prototyping the analysis, and generating the first version of the code.

#Troubleshooting

• Tools not listed: restart the server from your editor's MCP panel and complete sign-in.
• 401 errors with a token: check the header is exactly Authorization: Bearer YOUR_TOKEN with no quotes inside the value.
• Stale-looking numbers: ask the agent to print the observation date for each value. Financials update with reporting periods, not daily.
• Wrong company: have the agent resolve names with search_stocks first and confirm the symbol before pulling data.

Start with the MCP overview, connect one editor, and run Workflow 1. It shows you within a few minutes how your agent uses the tools and how many calls a typical task takes.