AI that reads your real wedding workspace

Anvaya AI lives on two surfaces that read and write the same wedding. The first is the chat on every dashboard page — it knows the context you are on. Open the guest list and ask “who hasn't RSVP'd to the sangeet?” and the in-app chat answers from the page you are already looking at. The second surface is the external MCP server at mcp.anvaya.love. Connect it to Claude, ChatGPT, or another MCP-capable client once over OAuth and the core Anvaya surfaces — guests, budget, vendors, RSVPs, tasks, events, seating — become callable tools inside that assistant.

Both surfaces share the same auth, role permissions, and audit log. A draft your mom started in the in-app chat picks up when she asks Claude to finish it from her phone. A budget line your partner added from another MCP-capable client shows up next morning in the dashboard with the same activity log entry. The data lives in one place; the surface is wherever you happen to be working.

The Model Context Protocol — the open standard Anthropic released in 2024 — makes the second surface possible. MCP gives AI assistants a structured way to call tools on remote servers. Anvaya uses that protocol to expose wedding planning tools and visual widgets through the assistants couples already use.

Most MCP servers return JSON and let the assistant describe it in words. Anvaya ships visual widgets — small React apps embedded as iframes — that render inside the Claude and ChatGPT conversation. Five widgets are shipped today: a dashboard with the wedding overview, a budget summary with variance against plan, a vendor compare with side-by-side quotes, an RSVP tracker with per-event state, and a task board organised by time horizon. Each widget is a real interactive surface — not a screenshot — that the assistant renders when the tool it called has a UI attached.

Widgets respect the Anvaya design language — cream background, burgundy primary, gold accents, Cormorant Garamond display. Inside ChatGPT or Claude the workspace looks like itself, not a JSON blob the assistant is paraphrasing. Best-value photographer ribbons in gold. Over-budget mandap lines glow red. The sangeet RSVP donut splits into confirmed, regrets, and pending arcs. A planner that looks like a wedding, not a Slack bot.

Mutating widgets run a two-stage confirmation. A single click never sends a message or charges a vendor — the widget transitions from idle to confirm to drafting to done, with a cancel path at every step. Read-only widgets render immediately; write actions wait for your tap.

A generic AI assistant will happily plan your “Indian wedding ceremony” as one event on a Saturday. Anvaya AI knows that a desi wedding is 3–5 days, 6 core ceremonies (with regional variations on top), and a guest list with overlapping cohorts that go to different events. The tools understand canonical South Asian ceremonies as first-class objects — mehndi, sangeet, haldi, baraat, pheras, walima, Anand Karaj, Garba, Gaye Holud — and pre-fill the right defaults per ceremony when you create one.

Cultural defaults run deep. Ask Anvaya AI to add a panditji and the record asks for the gotra and the muhurat window. Ask it to draft a haldi reminder and the message picks a morning time. Ask for a checklist and tasks come back tagged to the right ceremony — dhol player under sangeet, baraat horse under baraat, pheras muhurat under ceremony. The vocabulary is in the data model.

Multilingual works the same way. Anvaya AI replies in the language you write in — Hindi, Punjabi, Tamil, Telugu, Bengali, Gujarati, Marathi, Urdu (transliterated in Latin script). When drafting a message to relatives, the assistant offers the same message in their preferred language without re-asking. The cultural knowledge lives in every tool the AI can call, not just the chat.

Connecting Anvaya AI to Claude takes about a minute. Open Claude Desktop or claude.ai, go to the connectors menu, search for Anvaya, and click connect. A consent screen opens in a browser window with the Anvaya wordmark, the name of the wedding the assistant will get access to, the email you are signed in as, and the six scopes the connector will use. Read the scopes — they are written in plain English — and click authorise. The window closes and Claude is connected.

The same flow works in ChatGPT and MCP clients that support authenticated remote servers. Once authorised, the access token works across sessions on that assistant. Adding a second assistant creates a separate, independently revocable connection.

The integrations page shows every connected assistant with the date you connected, the scopes you granted, and the date of the last tool call. Revoking takes one click. No SaaS subscription is required for the AI tools themselves — you pay the assistant directly, on whatever plan you already use.