App

Starting the App

Launch the MCPLab web UI and find your way around.

Launch the App

The mcplab app command starts a local web server and opens the UI in your browser.

basic launch

npx @inspectr/mcplab app

with options

npx @inspectr/mcplab app \
  --evals-dir ./evals \
  --runs-dir ./runs \
  --libraries-dir ./libraries \
  --port 3000 \
  --open

Default Directories

When directory flags are omitted, MCPLab looks in the current working directory for eval configs and uses ~/.mcplab/runs for run output.

--evals-dir defaults to the current working directory.
--runs-dir defaults to ~/.mcplab/runs.
--libraries-dir is optional. No library is loaded if omitted.

Navigating the UI

The sidebar gives you access to all App features.

Dashboard — recent runs and pass rate trends at a glance.
Configurations — browse all eval configs grouped by suite, filter, and run per suite.
Run Evaluation — select a config and launch an eval.
Results — browse completed runs and drill into detail.
MCP Tool Analysis — review MCP tool definitions for quality.
Library — manage reusable agents and servers.
Settings — configure API keys, preferences, and view MCP Connection Info.

MCP Connection Info

The Settings page includes a Connection Info section that shows the MCP server endpoint URL for your running MCPLab instance. It also provides ready-to-paste configuration snippets for connecting external agents — including Claude, OpenAI-compatible clients, and other MCP-aware tools — to the MCPLab MCP server.