English · 한국어

@mcpeak/dashboard

Local web UI for MCPeak, the MCP server testing framework. Pick and run suites, and carry out generate's approval screen and repair review in the browser.

Install and run

npm install -g @mcpeak/dashboard
mcpeak-dashboard

Note: MCPeak's own messages are currently printed in Korean; English translations are shown alongside.

대시보드: http://localhost:7357  (스위트 탐색 루트: /path/to/project)

Dashboard: http://localhost:7357 (suite search root: /path/to/project)

It looks for suite files under the directory the command was run in. If the list is empty, check the search root printed at startup first. Change the port with --port; 0 picks a free port automatically.

Screens

| Screen | What it does | |---|---| | Home | The list of suites found. Fill in the server run command and options to run one | | Run | An event stream for the run in progress and per-case results | | Generate wizard | generate's target selection, dry run, and approval screen | | Repair review | Get and review cause candidates from a failed run's bundle | | Replay | External session record and replay | | Analyze | The tokens tab shows the compression report from Optimize, the security tab the findings from Audit | | Settings | Provider, model, and so on |

Verdicts match the CLI

The dashboard has no verdict logic of its own. It calls the command functions that @mcpeak/cli/commands exposes directly, and only swaps the terminal output and interactive prompts for web implementations. So the same suite gives the same result whether run through the CLI or the dashboard. If the verdict logic existed in two copies, the results would eventually diverge, and on that day the user would have no way to know which one to trust. Background: ADR-0046 (in Korean).

Requirements

Node 22.18 or later.

Documentation

Dashboard guide · MCPeak docs

License

MIT