MoSPI
by MoSPI
Query India's official government statistics from your AI agent. GDP, retail inflation, unemployment, industrial production, health, education, energy and environment data from 27 national datasets. 4 tools, all read-only, no sign-in.
Verified connector
Listed by Anthropic as a partner connector in its Connectors Directory.
Connection checked by Agentman on .
Anthropic states this reflects the level of review a connector received, not a security audit.
Connect MoSPI via MCP
https://mcp.mospi.gov.in/Works in any MCP-compatible client. In Agentman Studio it is one click — no config file to edit.
MoSPI Tools & Capabilities (4)
Returns the full list of available indicators for a given dataset. Datasets often have broader coverage than expected — for example, ASI covers 57 indicators (capital structure, wages, employment, GVA, fuel consumption), and GENDER covers 147 indicators across health, education, labor, and crime. For PLFS and ASUSE, indicators are grouped by frequency_code: - PLFS frequency_code=1 (Annual): all 8 indicators including wages - PLFS frequency_code=2 (Quarterly): indicators 1-3 only - PLFS frequency_code=3 (Monthly): indicators 1-3 only frequency_code selects the indicator set, not time granularity. Step 2 of: list_datasets → get_indicators → get_metadata → get_data
Returns the valid filter values (states, years, quarters, etc.) for a given dataset and indicator. Filter codes are arbitrary and dataset-specific — for example, PLFS state_code 99 means "All India", and NAS frequency_code 1 means "Annual". These values cannot be inferred or guessed from parameter names alone. The returned filter_values and api_params should be used as-is when calling get_data. Step 3 of: list_datasets → get_indicators → get_metadata → get_data
Fetches statistical data from a MoSPI dataset. This is the final step of the workflow. It requires filter values from get_metadata — filter codes are arbitrary (e.g., indicator_code=3 means "Unemployment Rate" in PLFS but something different in other datasets). All filter parameters including limit and page go inside the filters dict, not as top-level arguments. Step 4 of: list_datasets → get_indicators → get_metadata → get_data
Returns an overview of all MoSPI statistical datasets with descriptions and coverage. This is the starting point ΓÇö call this first to identify the right dataset. The API covers 500+ indicators across employment, prices, industry, national accounts, health, education, disability, housing, environment, trade, and more. Each dataset has its own indicator codes, filter parameters, and valid values ΓÇö these are not standardized and cannot be inferred or guessed from parameter names alone. Four-step workflow (each step depends on the previous): 1. list_datasets() ΓÇö identify the dataset 2. get_indicators(dataset) ΓÇö list available indicators 3. get_metadata(dataset, indicator_code) ΓÇö retrieve valid filter values 4. get_data(dataset, filters) ΓÇö fetch the data Returns: dict with 'datasets' (name, description, use_for for each dataset) and 'workflow' (the four-step sequence).
Read from the server on 2026-08-17, including each tool's own safety annotations.
Limits
- Aggregate statistics only — no microdata. Every tool hits MoSPI's indicator APIs, which return figures filtered by indicator, year and geography. No tool in the live handshake accepts a household, person or enterprise identifier, and no schema exposes a unit-level record endpoint. This is a sourced negative: we checked all four schemas, not just the tool names.
- A pilot project, not a certified feed. MoSPI's repository describes the work as a Model Context Protocol pilot undertaken by the ministry. The connector routes to published data; it does not stamp any figure as final.
- Figures may be provisional or superseded. The national accounts specification exposes a
revision_codewith seven values plus current and back series, so one year's GDP exists in several vintages. An answer that omits base year, series and revision is ambiguous, not wrong. - District data exists for two datasets only. The Economic Census returns ranked districts, and UDISE carries district-level school data. Everything else floors at state or national level.
- Economic Census state codes are not portable. Its specification states each round uses a different coding scheme, and that the fifth round does not support activity filtering at all.
- Results are paginated and default small. The PLFS specification notes the default of 10 records per page is "too low for most queries" and the
get_datadescription says pagination arguments belong inside the filters object, not at the top level. An agent that forgets this silently truncates its own answer. - Some indicator parameters reject multiple values. The PLFS specification warns that a comma-separated
indicator_codecauses a 500 error on every frequency code, though other filters accept comma-separated lists. - India only. Every dataset is a domestic instrument of India's National Statistical System. There are no international comparators.
- The ministry's own documentation page was unreachable from our network. See Sources.
- We did not exercise any tool. Our check was a read-only
initializeandtools/listhandshake. Everything above comes from the server's own tool descriptions, MoSPI's repository and the published API specifications — not from calling anything.
Frequently asked questions
Does the MoSPI MCP server need an account or API key?
No. We opened an anonymous handshake against the endpoint on 2026-08-17 and the server returned all four tool definitions without any credential. Anthropic's directory records the connector as authless with a no-auth posture. The server's own source code calls the underlying MoSPI data APIs with no key, token or registration step of any kind.
Which Indian datasets and surveys does the MoSPI connector cover?
The live get_indicators schema accepts exactly 27 dataset names, matching the total in MoSPI's own changelog. They include PLFS for employment, CPI and WPI for inflation, IIP and ISP for production, NAS for GDP, ASI for factories, plus NFHS, AISHE, UDISE, HCES, the Economic Census and seven NSS rounds. MoSPI's repository documents what each one covers.
Can the MoSPI MCP server return unit-level microdata about individuals?
No. Every tool queries MoSPI's published indicator APIs, which return aggregate statistics filtered by indicator, year and geography. No tool in the live handshake accepts a household, person or enterprise identifier, and no schema exposes a record-level endpoint. The lowest geography any dataset offers is the district, in the Economic Census and UDISE only.
Why must MoSPI tools be called in a fixed order?
Because the filter codes are arbitrary and cannot be guessed. MoSPI's repository states that skipping the metadata step produces invalid filter codes, and the live tool descriptions confirm that indicator_code=3 means unemployment rate in PLFS but something different elsewhere. Agents must walk list_datasets, get_indicators, get_metadata and then get_data in sequence.
How far back does MoSPI data go, and which base year applies?
Coverage varies by dataset and each carries its own base year. The published API specifications show annual PLFS running 2017-18 to 2023-24, UDISE covering 2018-19 to 2024-25, and CPI offering 2024, 2012 and 2010 base years. National accounts support 2022-23 and 2011-12, while the Economic Census reaches back to its fifth round in 2005.
Are the figures returned by the MoSPI connector official statistics?
They come from MoSPI's own eSankhyiki data APIs, and the ministry authors the connector. But the repository describes the project as a pilot, and national accounts expose a revision code with seven values, so a figure may be provisional. Treat the connector as a query route to published data, not as a certified publication channel.
Sources
- MoSPI eSankhyiki MCP Server repository,
nso-india/esankhyiki-mcp(retrieved 2026-08-17; HTTP 200). The GitHub organisation is India's National Statistical Office. Source for the dataset reference table, the 27-dataset changelog figure, the mandatory tool order, the MIT licence, the self-hosting and client setup paths, and the pilot description. · retrieved 2026-08-17 - MoSPI dataset API specifications, same repository,
swagger/directory (retrieved 2026-08-17; HTTP 200 each). Source for per-dataset time coverage, base years,state_codeanddistrict_codeavailability, the NASrevision_code, the Economic Census round-specific coding schemes, and the pagination and comma-separated-value constraints. · retrieved 2026-08-17 - Live
initializeandtools/listhandshake againsthttps://mcp.mospi.gov.in/— 4 tools with descriptions, input schemas and read-only annotations, returned to an anonymous request (2026-08-17). Source for the tool names, the four-step workflow, the dataset name enumeration, the 500+ indicator figure and the arbitrary-filter-code warning. · retrieved 2026-08-17 - Anthropic Connectors Directory entry (listed 2026-03-30; snapshot taken 2026-08-16; the directory app returns HTTP 403 to an unauthenticated fetcher, so the values below come from the committed snapshot rather than the rendered page). Source for the partner tier, the endpoint with its trailing slash, the read-only permission label, the ministry authorship, and the four stale
step1_know_about_mospi_apithroughstep4_get_datatool names. · retrieved 2026-03-30 - MoSPI privacy policy (retrieved 2026-08-17; HTTP 200, no redirects). · retrieved 2026-08-17
- eSankhyiki portal and MoSPI open API host — and (both retrieved 2026-08-17; HTTP 200, no redirects). Both render client-side and yielded no extractable text to our fetcher. · retrieved 2026-08-17
- Access note. The
documentationURL in Anthropic's directory entry,https://www.datainnovation.mospi.gov.in/mospi-mcp, was unreachable from our network on 2026-08-17: TCP connect to 103.208.110.25 on ports 443 and 80 timed out or was refused, from curl, from our fetcher and from a real browser session alike. Control hosts on adjacent MoSPI infrastructure —www.mospi.gov.in,esankhyiki.mospi.gov.inandapi.mospi.gov.in— all returned HTTP 200 to the same client at the same time, so this is a reachability finding about our egress path, not a claim that the ministry's documentation is down. Itsrobots.txtcould not be read for the same reason, so no page on that host was fetched or used. This page is therefore built from the live handshake, the directory snapshot and MoSPI's own public repository, which the ministry's documentation page links to as the project's source code. · retrieved 2026-08-17
Use in Agentman
Connect once and your agents call these tools on their own — on a schedule, in a workflow, with nobody at the keyboard.
Open in Agentman StudioServer Info
- Category
- Data & Research
- Developer
- MoSPI
- Tools
- 4
- Domain
- mcp.mospi.gov.in
Using Claude Desktop or another MCP client? Setup docs — the connection URL above works anywhere.