The connector layer for India's economy
Ship agents that read the source.
Not the coverage.
25 official Indian sources as MCP servers, behind one metered gateway — callable from the client your team already runs.
Evaluate it — no account required
Describe what you need in plain words. You get the connectors that carry it, the level they report at, the years they cover, and where the record stops.
Searches all 25 live connectors
Built on the primary record from
Published is not the same as accessible.
The figures sit in annexure tables inside PDFs, behind JavaScript dashboards, and behind portals that fingerprint and block collection. So a model asked about Indian inflation answers from a summary of a summary.
One protocol in. One gateway out.
No bespoke SDK to adopt. Every connector is a Model Context Protocol server, and every call it makes crosses the same five checks on the way through.
- Protocol
- Remote MCP over HTTP, behind OAuth 2.1 with PKCE and dynamic client registration. 5 first-party clients verified against it — Claude Code, Codex CLI, Gemini CLI and their SDKs — and any other MCP client connects with the same URL.
- Metering
- Every call routes through one gateway on your account, through a five-rung ladder, and every refusal is recorded by reason — so “nobody used it” is never confused with “everybody was turned away”.
- Provenance
- Each catalogue is harvested and enumerated, not described. What a source holds, at what grain, over what period, read on a dated pass — so a coverage claim is never a claim about the past pretending to be about the present.
- Delivery
- Assemble sources into an app and publish it as a link, an embedded widget, or a client config — with an origin allowlist and per-app limits attached to it.
The same machinery, pointed inward.
A connector is a harvest plan and a set of tools over material a machine cannot otherwise reach. Whose material it is turns out to be a detail.
India's record, opened up
The Reserve Bank's warehouse, national statistics, Union Budgets, parliamentary answers, vehicle registrations, GST collections, payment volumes — enumerated and given tools an agent can drive.
Then your own material
None of this is specific to government data — it is specific to material a machine cannot reach. Point it at your warehouse, your document store, your vector collections, or a source that publishes on a schedule, and an agent answers from your numbers beside the national ones, in one conversation.
Start reading, or start building.
In that order — and you can stop after the first.
Evaluate it in the browser
Describe what you need to work out. Find which sources answer it, at what grain, how far back — and where the record runs out. No account to read it.
Ask the catalogue →02Connect your own stack
One config puts these tools inside the client your team already runs. The same connectors you just tried, called from your own agents.
Configure a client →03Ship it to your users
Put several sources behind one set of instructions and publish — a link, a widget inside your product, or a schedule that runs without anyone watching.
Build an app →Start where it costs nothing
Ask it something you actually need to know.
No account to read the catalogue. You will know within a minute whether the record you need is in here, at the grain you need it, and how far back it goes.