Cloud Quickstart
Connect a coding agent to EigenVertex — no infrastructure required.
Connect a coding agent (Claude Code, Claude Desktop, Claude.ai, or any MCP-compatible assistant) to your own EigenVertex Corpus or Constellation, and have it build a working integration for you — directly from your project.
This page is for EigenVertex Cloud customers. If you’re self-hosting EigenVertex, see the operator documentation instead — ask your EigenVertex contact for access.
What you need first
- An EigenVertex account, with at least one Corpus (a Workspace with content in it) — create one in Studio if you don’t have one yet.
- Optionally, a Constellation grouping one or more Corpora — also created in Studio.
You do not need the eigen CLI, a database connection, or any infrastructure
credentials. Everything below works from a browser login and, for the code-integration path,
a single API key.
Option A — Live tools, right in the conversation (MCP)
This is the fastest way to get started: your coding agent gets live tools to inspect, query, and even write to your Corpus, right inside your conversation with it.
Claude Desktop or Claude.ai (web)
- Settings → Connectors → Add custom connector.
- Server URL:
https://api.eigenvertex.com/mcp - Click Connect — this opens EigenVertex’s login screen (Google, Apple, or email/password). No manual Client ID/Secret needed; EigenVertex registers the connection automatically.
Claude Code (terminal, inside your project)
claude mcp add --transport http eigenvertex https://api.eigenvertex.com/mcpThen, inside a Claude Code session in that project, run /mcp and follow the
browser authentication prompt.
What you can do once connected
Just ask, in plain language — the agent picks the right tool:
❭ Check EigenVertex’s status.
❭ List my corpora.
❭ Ask my “Piano Tuning Expert” Constellation what causes string inharmonicity.
Returns a real, cited answer grounded in your own content.
❭ Ingest this note into my [corpus name] corpus.
The agent can write content it produces directly into a Corpus.
❭ Search only my [corpus name] corpus for X.
Without pulling in other corpora, even if it belongs to a wider Constellation.
Some write actions ask for a one-time approval in the interface before running — that’s expected.
What this connection cannot do
Create or reveal raw API keys. That’s deliberate — key material never crosses this connection. Use Option B below for that.
Option B — Build the integration into your own app (server-side)
Use this when you want a chat feature (or any programmatic access) running inside an app you’re building — a SvelteKit site, a backend service, anything with its own server.
1. Get an API key
Studio → Account → API Keys → Create key. Choose the Workspace it applies to, give it a name, and copy the key — it’s shown once.
Current limitation
The Studio key form issues a combined read+write key; there is no separate read-only option from the UI yet. Treat the key as sensitive accordingly, and store it only in your server’s secret store — never in frontend code or a public repo.
2. Mount the shared chat relay (don’t hand-write one)
The relay — the server-side code that forwards to EigenVertex’s OpenAI-compatible endpoint while keeping your key off the browser and the target Constellation fixed — is a package EigenVertex ships, not something to build yourself.
from eigenvertex.integrations.fastapi import create_chat_relay_router
router = create_chat_relay_router(
base_url="https://api.eigenvertex.com",
api_key=EIGEN_API_KEY, # from your server’s environment, never the client
constellation_id=YOUR_CONSTELLATION_ID,
)
app.include_router(router)Your frontend calls the mounted route (e.g. /chat) — it never sees the key or the
constellation ID.
3. Add a chat client
Point any chat UI component at your mounted route. If you’re using an AI coding assistant to build this, just describe what you want — e.g. “Add a chat panel to this app, backed by our relay endpoint at /chat.”
Core concepts, briefly
A Corpus can belong to at most one Constellation today.
Known limitations, stated plainly
- No separate read-only key from Studio’s UI yet — full read+write only, for now.
- A Corpus can only feed one Constellation at a time.
- Streaming responses arrive as a complete answer sent in one burst over MCP tool calls (not token-by-token); the server-side relay (Option B) does stream token-by-token to your frontend.
Want more detail?
This page covers the essentials. The full guide adds a complete worked example with a real terminal screenshot, the full catalogue of tools available over MCP, the security model behind cross-account isolation, and a troubleshooting FAQ.