Your data questions don’t always start in a chat window. Sometimes they start in your terminal, mid-task, inside Claude Code or Codex. As of this release, those agents can talk to GenViz directly: the desktop app embeds a Model Context Protocol server, so any MCP-capable agent on your machine can work with your data through GenViz — with your permissions, your connections, and full visibility in the app.
What agents can do
- Discover everything. List your workspaces, environments, and conversations, read message history, and see which models are available.
- Ask real questions. An agent can open a conversation in any environment and ask in plain language. GenViz runs the full pipeline — schema inspection, queries against your databases, streaming the answer back token by token through MCP progress notifications.
- Pull the data itself. When a question produces an extract, the agent can page through the rows as JSON or download the parquet file to local disk and analyze it with DuckDB, pandas, or anything else.
Everything stays visible
Agent conversations are first-class conversations. They show up in your sidebar with a bot badge showing which agent asked (“Asked by claude-code”), stream live if you have the conversation open, and file themselves into the Settled group when the agent’s turn completes — ask a follow-up and they pop right back.
Connecting an agent
While the GenViz app is running, it serves the MCP endpoint at http://127.0.0.1:9008/mcp. The first time it starts, it also creates an
access token file you’ll reference once during setup — you never need to
open or copy it:
| OS | Token file |
|---|---|
| macOS | ~/Library/Application Support/GenViz/mcp-auth-token |
| Windows | %APPDATA%\GenViz\mcp-auth-token |
| Linux | ~/.config/GenViz/mcp-auth-token |
For Claude Code, one command reads the token for you and registers GenViz. On macOS:
claude mcp add --transport http genviz http://127.0.0.1:9008/mcp --header "Authorization: Bearer $(cat "$HOME/Library/Application Support/GenViz/mcp-auth-token")"
On Linux:
claude mcp add --transport http genviz http://127.0.0.1:9008/mcp --header "Authorization: Bearer $(cat "${XDG_CONFIG_HOME:-$HOME/.config}/GenViz/mcp-auth-token")"
On Windows (PowerShell):
claude mcp add --transport http genviz http://127.0.0.1:9008/mcp --header "Authorization: Bearer $(Get-Content "$env:APPDATA\GenViz\mcp-auth-token")"
Prefer not to hand-edit paths? The app writes these commands for your exact machine: open your profile menu and choose Help → Set up MCP to copy the ready-made command for Claude Code or Codex.
Start a new Claude Code session and the GenViz tools are available. The token only guards the local port — agents act through your signed-in GenViz session, see only what you can see, and lose access the moment you sign out.
What’s next
Model-aware asks, extract analysis recipes, and deeper agent workflows are on the way. If you build something with the bridge — or want more from it — tell us at support@genviz.ai. Feedback on what your agents should be able to do next shapes what we ship.