DeepInsights is a connector layer: a set of Model Context Protocol servers over India's official economic record, plus connectors you can point at your own material. Agents call them directly. Nothing here is a download.
There are 22 connectors exposing 393 tools across 11 domains — the Reserve Bank's warehouse, national statistics, Union Budgets, parliamentary answers, vehicle registrations, GST collections, payment volumes, and more.
Ask, in the browser. Describe what you are trying to work out and the catalogue consultant tells you which sources can answer it, at what grain, how far back, and where the record runs out. It does not report figures — see what the consultant will not do below.
Call them from your own agent. Every connector speaks remote MCP over HTTP, so the client your team already runs can use them. See Connect a client.
Assemble an app and ship it. Put several sources behind one set of instructions and publish it as a link, a widget inside your own product, or a client config. See Build an app and Publish & embed.
your agent ─┐
├─► gateway ─► connector ─► source
an app ─────┤ (auth, (MCP server) (RBI, MoSPI, …)
│ metering)
a widget ───┘
Every path goes through one gateway. That is what makes usage countable and limits enforceable no matter which door a call arrives through — an agent you configured, an app you published, or a widget on someone else's page.
An MCP server with a harvested catalogue behind it. The distinction matters when you are deciding whether a source can answer your question:
| | | | --- | --- | | Enumerated | We have read what is inside — datasets, indicators, the grain, the period. Coverage claims about it are specific and dated. | | Described | The server does something rather than holding something (research, extraction, visualisation). It has no listable contents, and we never present it as though it does. |
Every connector page shows which it is, along with a harvest date. A coverage claim without a date is a claim about the past pretending to be about the present.
It never states a value, a total, a share, a ranking or a trend. It has not read the data — it has read the catalogue. Its job is to establish what can be answered, by what, at what grain.
To get actual figures, call the connector: from your own agent, from a playground, or from an app you build. That is the thing that reads the data.