VecViz Documentation
Connecting to VecViz
VecViz exposes its model as an MCP server, so any MCP-capable agent can query it directly. There are two ways in.
Direct
The shorter path, but it requires a client that accepts an API key in its configuration. Claude Code, Cursor, VS Code, Gemini CLI and Windsurf all qualify.
Through Smithery
Works for every client, including Claude Desktop, claude.ai and the Claude mobile apps, which accept only OAuth connectors. Smithery holds your key and presents an OAuth-capable endpoint on your behalf.
Either way you need a VecViz API key first.
Get an API Key
- Free trial Ten tickers (AAPL, GLD, JPM, MRK, MSFT, SPY, TLT, TSLA, WMT, XOM), one click, no card, at agents-vecviz.fly.dev/try
- Full access The whole 788-ticker universe, $39 per month, cancel anytime, at vecviz.com/signup
Your key starts with vv_. Treat it the way you would any credential.
Direct Connection
The endpoint is https://agents-vecviz.fly.dev/mcp/, authenticated with your key as a bearer token.
Claude Code
claude mcp add --transport http vecviz "https://agents-vecviz.fly.dev/mcp/" \
--header "Authorization: Bearer vv_your_key_here"
Add --scope user to make it available across all your projects rather than just the current one.
Cursor, VS Code, Windsurf and Other Config-File Clients
{
"mcpServers": {
"vecviz": {
"type": "http",
"url": "https://agents-vecviz.fly.dev/mcp/",
"headers": {
"Authorization": "Bearer vv_your_key_here"
}
}
}
}
The file location differs by client. Cursor uses ~/.cursor/mcp.json; VS Code and Windsurf have their own paths. Restart the client after editing.
Just Ask Your Agent
Most coding agents can wire this up themselves. Paste this:
Connect to the VecViz MCP server at https://agents-vecviz.fly.dev/mcp/ using bearer authentication with the API key vv_your_key_here, then call
vecviz_catalogto confirm the connection.
Connecting Through Smithery
Use this route for Claude Desktop, claude.ai and Claude mobile, or for any client that will not take a raw API key. It takes a few more steps because you are provisioning your own proxy endpoint, not using a shared one.
- Create a Smithery account at smithery.ai. Note the namespace it gives you. It is usually your username, and it becomes part of your connection URL.
- Open the VecViz listing at smithery.ai/servers/vecviz/vecviz.
- Click “Add to toolbox” and pick the namespace you want the connection to live under.
- Click “Set up” on the VecViz entry that appears in your Toolbox, and paste in your
vv_key. The connection stays inactive until you do this. - Click “Install” and choose your client from the list.
Claude Desktop, claude.ai and Claude Mobile
Smithery gives you a URL of the form https://mcp.smithery.ai/your-namespace. In Claude, go to Settings > Connectors, click Add custom connector, and paste it in.
Custom connectors are not available on every claude.ai plan. If you do not see the option, check your subscription tier.
Everything Else
Smithery’s install dialog covers Claude Code, Codex, Cursor, VS Code, Gemini CLI, LM Studio, Cline, Zed, OpenCode, Antigravity, Cherry Studio, ChatWise, Roo Code, Amazon Q, Dust, LibreChat, Kiro and others. Pick yours and follow the command or config it produces.
For a client not on the list, the “Advanced” panel gives you raw JSON. That config authenticates to Smithery with a Smithery API key, not your VecViz key. Your VecViz key is already stored on the connection itself from step 4.
Two Things Worth Knowing
Your namespace URL is personal. It is bound to your toolbox and your stored key, so anyone you shared it with would be spending your quota. It is not a link to publish or pass around.
Your toolbox bundles every server you have added to that namespace behind one URL, so if you add others alongside VecViz, they all arrive together and tool names may come back prefixed with the connection ID.
Confirming It Works
Ask your agent to call vecviz_catalog. It is free and returns the coverage list, which tells you the transport is up.
Then ask for vecviz_read on SPY. That one is metered, so a clean response confirms your key made it through and is being counted against the right account. If vecviz_catalog succeeds but vecviz_read fails, the connection is fine and the problem is the key.
From there, methodology explains what any number means, and model_review shows the published track record. Both are free.
Where to Start
vecviz_read on a single ticker is self-sufficient: target price, channels, VecEvents, V-Score and which price-probability bands exist. Most sessions start there and follow the thread into vecevent_matrix to see why the target is what it is, or price_likelihood to ask whether a given price is reachable.