Updated 2026-08-20, checked against the live endpoint. KundaliMCP has consolidated its surface to nine product-level tools: kundali, lifemap, kundali_milan, shubh_muhurat, panchang, pramaan, chat, submit_feedback, get_version. It offers single-school Parashari (BPHS) interpretation in seven languages, with a 409-yoga catalog. It embeds Vedaksha 6.0.1, so it does not necessarily track the engine version this site reports. The previous version of this note said 17 tools, which is Vedaksha's engine-primitive count and was never the product's: the two are different facts. The launch narrative below is otherwise unchanged.
When we shipped Vedaksha v2.0.0, people kept asking who uses an astronomical engine. Nobody did yet. A clean-room ephemeris is most useful when something is built on top of it.
That something is KundaliMCP, a Vedic astrology product for AI agents. It launched this week. Both Vedaksha and KundaliMCP are built and operated by ArthIQ Labs, so this is a maintainer-built reference deployment rather than a third-party endorsement. We run our own engine in production, which we think is the right way to prove that it works.
What KundaliMCP is
KundaliMCP is a set of Jyotish tools served over the Model Context Protocol. An AI agent points at the server and gets a small set of composed tools: cast a chart, qualify yogas, forecast dasha periods and events, search auspicious windows, query divisional charts, check compatibility, recommend remedies, generate localized narratives, and emit graph representations for downstream RAG pipelines.
Interpretation follows the Parashari (BPHS) tradition. Output is offered in seven languages (English, Hindi, Sanskrit (Devanagari), Tamil, Telugu, Kannada, and Bengali), picked up directly from the Vedaksha locale crate.
KundaliMCP stores no personally identifiable information. Birth data goes in, computed charts come out, and nothing in between is persisted. This works because Vedaksha is a stateless engine: every chart is a pure function of (julian day, latitude, longitude, configuration). The product layer can be PII-blind because the engine layer never had a session to begin with.
What we kept from the engine
Every astronomical computation in KundaliMCP comes straight from Vedaksha, unchanged. The AnalyticalProvider with VSOP87A planetary series and ELP/MPP02 lunar theory runs in production untouched. The same ayanamsha implementations are exposed to end users: whatever the engine ships, the product ships. The same osculating-node calculation that agrees with JPL DE441 to within 0.6 arcseconds places Rahu and Ketu in every chart the product returns.
When a customer reports a position that disagrees with another tool by a few arcseconds, we trace it to a specific ayanamsha definition or precession model, the same way we always have, because the engine is the same. The product layer reimplements nothing, so there is no drift between what the docs say and what production runs.
What the product layer added
Vedaksha exposes 17 MCP tools, which are the engine primitives. KundaliMCP layers product-level compositions on top:
None of these belong in the engine, because they encode product opinions. All of them are short compositions of Vedaksha primitives. The architecture was designed for a thin product layer over a complete engine.
What we learned about MCP-first product surfaces
Building an MCP-first product took less work than we expected, because the engine already speaks the protocol. Vedaksha's MCP server implementation handles JSON-RPC 2.0, schema generation, validation, and error semantics. KundaliMCP inherits all of it, then layers product-specific tools on top using the same tool-definition pattern. There is no REST adapter and no GraphQL schema to maintain; MCP is the API.
The trade-off is that the audience is narrower, for now. MCP-aware AI agents can use KundaliMCP today; legacy clients need a thin REST shim. We decided to treat the agent surface as the default and let non-MCP clients adapt to it.
What this means for Vedaksha
An engine with one production user is not an ecosystem. KundaliMCP shows something narrower: the engine carries a full product without modification. Its performance, accuracy, type ergonomics, locale coverage and MCP semantics are now tested under real production query load.
If you are considering building an astrology product, an AI-agent assistant, or a research tool that touches planetary positions, the engine you would use is the same one running at kundalimcp.com right now. We will keep it that way.
Try it: kundalimcp.com · Build on it: cargo add vedaksha · Read the docs: /docs