MCP for coding agents

Expose your kinora test data to Claude Code, Cursor, and other agents over MCP.

@kinora/mcp is an MCP server that exposes your kinora test data to coding agents (Claude Code, Cursor, Claude Desktop, Windsurf, and others). Point an agent at your last CI failures and let it pull the error, the trace, and the flaky-vs-regression history to go straight to the fix.

It runs locally over stdio and talks to either the hosted cloud or your self-hosted server.

Setup

Add it to your agent’s MCP config. It needs a kinora API token (create one in the dashboard under project settings) and, for self-host, your server URL.

{
  "mcpServers": {
    "kinora": {
      "command": "npx",
      "args": ["-y", "@kinora/mcp"],
      "env": {
        "KINORA_TOKEN": "<your api token>",
        "KINORA_URL": "https://api.kinora.dev"
      }
    }
  }
}

KINORA_URL defaults to the hosted cloud; set it to your own origin for self-host. Both can also be passed as --token / --url flags.

Tools

Tool What it returns
list_projects Every project with its latest run summary.
list_failures Failing/flaky tests of a run (defaults to latest): error, file:line, trace URL.
get_run Full report for one run: counts, git/CI metadata, all test statuses, failures.
test_history Per-test history: health (stable, flaky or broken, see Flaky and broken tests) with its reason, pass/fail/flaky rates, and newlyBroken/newlyFlaky flags to tell a fresh regression from a chronic/flaky test.
get_trace The Playwright trace.zip URL for one test, to open in the viewer or download.
component_health Per-component and per-story health from component tests and Storybook stories: latest status, health over the last 20 runs with its reason, accessibility violations from Storybook’s a11y addon, and the tests mounting each story. Filter by component, or by health (flaky or broken).