MoSPI MCP Server

What is MoSPI MCP Server?

MoSPI MCP Server is the official hosted endpoint (https://mcp.mospi.gov.in/) from India's Ministry of Statistics and Programme Implementation that lets AI assistants query official Indian government statistics such as GDP, inflation, employment and industrial output. It is free and needs no sign-in.

MoSPI, through the National Statistics Office (NSO), publishes India's core economic and social statistics on its eSankhyiki portal and open data APIs. The MCP server is a bridge between those APIs and assistants like Claude, ChatGPT and Cursor, so economists, journalists, students, policy researchers and developers can ask for a statistic in plain language instead of navigating survey tables and filter codes by hand. The server code is open source under the MIT license in the nso-india/esankhyiki-mcp repository.

How to use MoSPI MCP Server?

No account or API key is required. Add the server URL to your client:

https://mcp.mospi.gov.in/

Claude Code

From the project README:

claude mcp add esankhyiki-mcp --transport http https://mcp.mospi.gov.in/

Verify with claude mcp list.

Claude Desktop and claude.ai

Open Settings, go to Connectors, choose "Add custom connector", name it MoSPI and paste https://mcp.mospi.gov.in/. No sign-in step follows.

Cursor

{
  "mcpServers": {
    "esankhyiki-mcp": {
      "url": "https://mcp.mospi.gov.in/"
    }
  }
}

The README's own Cursor and Windsurf example wraps the URL with npx mcp-remote https://mcp.mospi.gov.in/, which also works if your client version only supports local commands.

VS Code

{
  "servers": {
    "esankhyiki-mcp": {
      "type": "http",
      "url": "https://mcp.mospi.gov.in/"
    }
  }
}

Self-hosting

The README also explains how to run the same server yourself with Python and FastMCP (python mospi_server.py, serving at http://localhost:8000/mcp) or with Docker.

Key features of MoSPI MCP Server

• Official source: run by MoSPI with code published by the NSO India organization on GitHub.
• Free, public endpoint with no authentication.
• Broad coverage: the tools describe more than 500 indicators across employment, prices, industry, national accounts, health, education, disability, housing, environment, trade and more.
• Major datasets include PLFS (labour force), CPI and WPI (prices), IIP and ISP (industrial and services production), ASI (factories), NAS (GDP and national accounts), NFHS (family health), energy, environment and RBI statistics.
• A guided four-step workflow that keeps the model from guessing dataset-specific filter codes.
• Open-source, self-hostable server with Docker support and OpenTelemetry observability.

MoSPI MCP Server tools

The server exposes four tools meant to be called in order:

• Find the dataset: list_datasets returns every available MoSPI dataset with a description and what it is used for. The agent calls it first.
• Find the indicator: get_indicators lists the indicators available in a chosen dataset.
• Find valid filters: get_metadata returns the valid filter values (states, years, quarters and so on) for a dataset and indicator. Filter codes are dataset-specific; for example, in PLFS a state code of 99 means All India.
• Fetch numbers: get_data retrieves the statistics using the filters from get_metadata.

Use cases of MoSPI MCP Server

• Economic research: "What was India's real GDP growth rate for the last five financial years?"
• Inflation tracking: "Show CPI inflation for food items in Kerala versus the all-India figure."
• Labour market analysis: "What is the unemployment rate for urban women according to the latest PLFS?"
• Journalism and fact-checking: confirm a quoted statistic against the official MoSPI source.
• Education and health studies: pull school enrolment from UDISE+ or nutrition indicators from NFHS.
• App development: give a data-analysis agent a reliable source of Indian official statistics.

Troubleshooting MoSPI MCP Server

• The model returns wrong or empty data: make sure it follows the full sequence list_datasets, get_indicators, get_metadata, get_data. Skipping get_metadata usually leads to invalid filter codes.
• Filters seem ignored: in get_data, all filter parameters, including limit and page, go inside the filters object, not as top-level arguments.
• A sub-survey is missing from the list: some NSS rounds have sub-parts (for example NSS77A or NSS79C) that are reached through their parent dataset, such as NSS77 or NSS79.
• Client cannot connect over HTTP: use the npx mcp-remote wrapper shown in the README.