Skip to main content
← CLI Reference

travsr ui

Browse the code graph in a web browser. It serves the same pages as the VS Code extension, so you get them without an editor: repo status, the repo map and symbol graph, callers, dependencies, blast radius, execution paths, the architecture brief, the context an agent would be handed, and a terminal for read-only CLI commands.

travsr ui --open

Everything is compiled into the travsr binary. There is nothing to install and no build step. The server runs inside the CLI, not the daemon, so the daemon gains no network surface.

Options​

FlagDefaultMeaning
--host <ADDR>127.0.0.1Address to bind: an IP address, or localhost. Must be a loopback address unless --allow-remote is also given
--port <PORT>7330TCP port to serve on. 0 picks a free one
--allow-remoteoffAllow binding a non-loopback address. The page has no authentication, so anyone who can reach the port can read the graph and the source of every file in the index. It also turns the action buttons off
--openoffOpen the page in your default browser once the server is up

The server runs until you press Ctrl-C.

Pages​

PageWhat it shows
HealthIndex status, daemon state, language readiness and the travsr version
GraphThe symbol graph, with symbol suggestions under the search box
Callers, Dependencies, Blast radiusThe matching MCP tool's answer for a symbol or file, with suggestions as you type
Execution pathThe lowest-cost call chain from one symbol to another, as get_execution_path returns it
ArchitectureThe facts agents get for explaining how the repo fits together, from get_architecture_brief
ContextWhat an agent would be handed for a question, as get_context returns it
TerminalA fixed set of read-only CLI commands, such as travsr ask and travsr graph, run exactly as typed in a shell with their own output shown. It never accepts raw arguments
ReposThe registered repositories

Open pages redraw when the index changes. Graph answers are the MCP tools' own text, so the page and an agent asking the same question see the same result.

What it can change​

Browsing is read-only. The page asks the daemon the same way the CLI does, and when no daemon is running it opens the index directly, always read-only: an index that needs migration is reported, not migrated.

The action buttons are the exception. They run daemon, reindex, fsck and connect commands from the page, and each one waits for a confirmation in the page first. --allow-remote turns them off.

Network exposure​

By default the server binds loopback only. While bound to loopback it also rejects requests whose Host header is not a loopback name, which stops a web page on another origin from reaching it by DNS rebinding. Responses carry a strict Content Security Policy: scripts only from the page's own origin, no inline script, never framed.

Treat --allow-remote as exposing your source tree on the network.

See also​