NSE-BSE MCP Server: Real-time NSE, BSE Data for AI

Why “NSE MCP server” is showing up in dev circles

Model Context Protocol (MCP) is being discussed as a standard way to connect AI models to external tools and data sources. Social posts describe MCP as an open standard designed for AI assistants like Claude, ChatGPT, and Gemini. In the Indian markets context, the key idea is simple: expose stock-market endpoints as “tools” an assistant can call. That approach is being framed as a cleaner alternative to ad-hoc scraping or one-off integrations. The focus in these threads is not trading, but reliable access to quotes, history, and exchange documents. Several projects position themselves as “MCP servers” for NSE and BSE data in one place. The common value proposition is tool discovery plus a consistent calling pattern through MCP. The discussion also highlights that multiple implementations exist, each with different dependencies and transports.

What the NSE-BSE MCP Server claims to provide

One widely shared project is branded as an “NSE-BSE MCP Server” for NSE and BSE India stock market APIs. Posts state it provides a Model Context Protocol endpoint to query real-time market data and related documents. It is described as supporting Streamable HTTP transport and being suitable for production environments. The feature list mentioned includes health checks, CORS support for browsers, and HTTP stream transmission. Several write-ups repeatedly cite “66 tools” as the headline capability. The same descriptions split those tools into 44 NSE tools, 20 BSE tools, and 2 document tools. The server is also described as open source and free for self-deployment and customization. A separate listing describes “NSE-BSE India Stock Market” with provider name “bshada”, released on Dec 1, 2025, and a rating shown as 2.5 points.

Tool coverage developers highlight most often

The tool lists shared on social media focus on day-to-day market data first. Examples include nse_get_market_status for trading hours and market status. Real-time quotes appear as nse_equity_quote and nse_get_quote, with the latter supporting segment specification like equities, SME, mutual funds, and debt. Symbol discovery is covered with nse_lookup_symbol, which searches by name or partial match. Movers show up as nse_get_gainers and nse_get_losers in the NSE tool set. Broader descriptions add options chains, corporate actions, and IPO information as part of the overall capability. Some posts also mention “trending stocks and price shockers” as covered categories. The emphasis across these discussions is breadth of tool coverage, rather than any single endpoint.

How data is delivered to AI clients

Transport and client support is a recurring topic, because it affects how quickly teams can integrate. The NSE-BSE MCP Server is described as using Streamable HTTP transport and being stateless. Social posts also claim it supports mainstream AI assistant platforms such as Claude Desktop, ChatGPT, and Gemini CLI. A separate Python-based “Indian Stock Exchange MCP Server” is described as an HTTP JSON-RPC server compatible with Claude Desktop, Cursor, and VSCode. In that Python project, the URL pattern shared for clients is http://localhost:8000/jsonrpc. The same documentation also notes JSON-RPC 2.0 with Content-Type: application/json for web clients. Some write-ups mention Registry Broker discovery resources and an MCP protocol hub for integration guidance. Overall, the market conversation is as much about wiring and transport as it is about the data itself.

Quick-start approaches shared by builders

The TypeScript-based NSE-BSE MCP Server is shown being started via npx nse-bse-mcp in multiple posts. Another path shared is running from source using Bun, with steps like bun install, bun run build, and bun start. These examples appear in “Quick Start” sections designed for fast local testing. The same server is described as having “smart response limiting” where large responses return metadata so the model can decide what to fetch next. That feature is framed as practical when an assistant might otherwise request too much data at once. The Python “Live-NSE-BSE-MCP” project describes starting via python -m ise_mcp.server after installing dependencies. It explicitly notes that a PyPI console script (ise-mcp-server) is not available yet, and recommends module execution or a batch script. Most examples assume local development first, and then connecting a desktop client or IDE plugin.

Document and research workflows beyond prices

Document handling is part of why some teams are evaluating MCP servers instead of basic quote APIs. The NSE-BSE MCP Server is described as including a “Document Downloader” that can download and extract IPO prospectuses, reports, and circulars. Posts also describe “document processing” as a supported capability for financial analysis and investment research. On the Python IndianAPI-powered server, the published tool list includes research actions such as search_industry, get_analyst_recommendations, and get_stock_forecasts. It also includes get_historical_data with various filters, alongside market lists like get_trending_stocks. Another commonly cited endpoint is get_nse_most_active, framed around volume-based activity. Separate descriptions mention “financial news and announcements” as part of market coverage. The pattern across projects is a combined workflow: fetch data, fetch documents, then let the assistant summarise or analyse.

A practical comparison of the MCP options mentioned

Teams comparing these servers are mostly weighing tool breadth, dependency stack, and data source requirements. Some projects describe direct integration with NSE and BSE stock market APIs, while others clearly depend on third parties. The IndianAPI-powered Python server requires an API key and will not start without ISE_API_KEY. The yfinance-based “kai-stock-market-mcp” positions itself as covering NSE and BSE via .NS and .BO symbols, while also supporting global tickers. Another project is an NSE ticker MCP server that accesses NSE data through the Upstox API and supports multiple time intervals. The table below summarises the concrete details that were explicitly shared in posts and READMEs. Where a detail was not specified in the shared context, it is left out.

Integration checklist that keeps coming up

Most integration guides start with choosing the MCP client, because setup differs between desktop apps, IDEs, and CLI tools. For HTTP JSON-RPC setups, the Python server documentation repeatedly points to the /jsonrpc endpoint and JSON-RPC 2.0 requests. For Streamable HTTP, the emphasis is on stateless calls and browser compatibility through CORS support. Environment variables are a common theme, especially on the IndianAPI-powered server where ISE_API_KEY is mandatory. That same guide also advises adding the .env file to .gitignore to avoid committing secrets. For local testing, the examples suggest starting the server first and then restarting the AI client so it picks up the MCP configuration. Several posts recommend testing with simple prompts like “top gainers today” or “data for Reliance” to confirm tool wiring. Teams also discuss response sizing, where “smart response limiting” is presented as a way to control large payloads. The recurring takeaway is that MCP reduces glue code, but you still need to be deliberate about auth, payload size, and tool selection.

What to watch before using it in production workflows

Developers in these threads treat “production ready” claims as a starting point, not a guarantee. One reason is that the ecosystem includes multiple servers with different upstream data sources and constraints. If you use a paid API like IndianAPI or a broker API like Upstox, key management and rate limits become part of your system design. If you use yfinance-based tooling, reliability depends on that upstream feed and symbol conventions like .NS and .BO. The listings also show a visible rating for at least one package, which prompts some users to test locally before deeper integration. Document download and extraction features can expand scope quickly, so teams discuss using metadata-first responses when possible. Client compatibility matters too, because some tools target Claude Desktop and Cursor, while others mention Gemini CLI and ChatGPT. The practical approach described is incremental adoption: start with quotes and symbol lookup, then add history, movers, and documents. Within the shared context, the most consistent guidance is to validate endpoints, auth requirements, and client configuration early.