WebMCP · in-browser AI agents · Gemini in Chrome · agentic browsers
Let your browser's AI agent compute your Vedic chart
A new browser standard, WebMCP, lets a website hand its features to the AI agent running inside your browser as real, callable tools. Eternal Evals is the first Vedic astrology engine to support it — so when an in-browser agent is on our pages, it computes your actual sidereal chart instead of reading and guessing at the screen.
What WebMCP is
WebMCP is a browser API — navigator.modelContext — that lets a web page register structured tools for AI agents to call. It's a W3C Community Group proposal from engineers at Google and Microsoft, available today in a Chrome origin trial (Chrome 149+) and natively in Edge 147+. The browser takes the tools a page registers and exposes them to any agent it hosts, translating them into MCP — the Model Context Protocol, the same open standard used by Claude and Cursor.
The practical shift is this: instead of an agent screen-scraping a page's HTML and inferring what it means, the agent reads a small registry of typed tools, picks the right one for what you asked, calls it with structured arguments, and gets a structured result back. No guessing at layout, no fabricated values.
Why it matters for astrology specifically
Most AI horoscopes are guesswork — the model has your birth date and fills in the rest from training patterns, and an agent left to read a generic astrology page can only lift prose that was never computed. Eternal Evals is built the opposite way. A deterministic engine (Swiss Ephemeris DE441, Lahiri ayanamsa) computes your chart first — planets, houses, 16 divisional charts, dashas, shadbala, yogas — and only then is anything interpreted. WebMCP extends that guarantee to the agent in your browser: it receives computed numbers, so it cannot invent a placement or a date. It reads them.
How to use it — there's nothing to install
If your browser has a WebMCP-capable agent, our tools are simply there when you visit an Eternal Evals tools or panchang page. Three ways that happens today:
1. Gemini in Chrome or an agentic browser — open a page like our Manglik calculator and ask, and the agent can call the tool directly.
2. Edge 147+ — WebMCP ships natively, no setup.
3. The Model Context Tool Inspector Chrome extension — a developer tool that lists and runs a page's registered tools, useful for testing.
On any tools page the agent sees the full catalogue — compute_vedic_chart, read_chart_section, and a read_… tool for every calculator (Manglik, Sade Sati, Navamsa, current dasha, Shadbala, Ashtakavarga and more). On the panchang page it also gets get_panchanga and find_muhurta. And because the tools run in the page with your own session, a signed-in visitor's agent can read their own saved charts — no API key, no copy-paste.
WebMCP vs. a remote MCP server
These are two doors into the same engine, for two different situations:
- Remote MCP server (
mcp.eternalevals.com) — for agents outside the browser: Claude Desktop, Cursor, Cline. You add it once as a connector. See the MCP setup guide. - WebMCP — for the agent inside your browser, active automatically while you're on our site, using your own session.
Building your own app instead of using an agent? The same engine is a plain REST API — see the developers page and /openapi.json.
An honest note on where this is
WebMCP is new. It's a proposal in a Chrome origin trial, and the population of browsers with a WebMCP-capable agent is still small and growing as agentic browsing rolls out. We've shipped it early on purpose: as agents increasingly operate the web rather than just read it, the sites they can actually use will be the ones that offered real tools. A computation-first engine is exactly the kind of thing an agent should be calling, not scraping — so this is a natural fit, and we'd rather be first than wait.