Skip to content

qlaneai/qlane

v0.7.1MIT

AI QA that runs your app in a browser on every pull request: set up a test target, read what it found on a pull request, and plan tests for a local diff.

Qlane agent plugin

Installs Qlane MCP into your AI coding editor, so your agent can reach your Qlane projects, test targets and what Qlane found on your pull requests without leaving the editor.

One server, every region

Qlane keeps each organization's data in one region, EU or US. The plugin registers one server for both:

ServerEndpoint
qlanehttps://mcp.qlane.ai/mcp

You sign in once and choose your organization. Each request then goes to the region that hosts it. There is no region to pick.

Until you sign in, the server contributes none of its own tools: your editor shows it with two sign-in helpers and nothing else. That is not an error state.

It is a plain static URL, so this works identically on every editor. Nothing is configured, substituted, or typed.

More than one organization

The server is connected to one organization at a time: the one you chose when you signed in. To switch, sign out of qlane.ai, then authenticate qlane again and choose the other organization. To stay connected to organizations in both regions at once, add the regional URLs directly, https://mcp-eu.qlane.ai/api/mcp and https://mcp-us.qlane.ai/api/mcp. They keep working.

Updating from an earlier version

Earlier versions registered two servers, qlane-eu and qlane-us. The update replaces them with qlane, so you authenticate once more after updating.

If you had also added a regional URL by hand with claude mcp add, Claude Code now shows every Qlane tool twice. Remove the server you added by hand. The earlier instructions named it qlane:

claude mcp logout qlane
claude mcp remove qlane

Install

Qlane MCP requires a Qlane account. Your editor prompts you to sign in the first time it calls the server.

Claude Code

claude plugin marketplace add qlaneai/agent-plugin
claude plugin install qlane@qlane-plugin

Then run /mcp and authenticate qlane. There is no configuration step and nothing to type.

Depending on your permission settings, Claude Code may ask before each Qlane tool call and each git, GitHub CLI or curl command a skill runs. To skip those prompts, allow them in your own Claude Code permission settings.

VS Code and GitHub Copilot

Add this repo to chat.plugins.marketplaces:

{
  "chat.plugins.marketplaces": ["qlaneai/agent-plugin"]
}

If nothing appears after adding it, check that chat.plugins.enabled is true.

Cursor

In Cursor's Import Marketplace dialog, enter the full repository URL, https://github.com/qlaneai/agent-plugin — Cursor rejects the qlaneai/agent-plugin shorthand there. Then install qlane and authenticate the qlane server.

Cursor also loads the plugin's .cursor-plugin/plugin.json, which gives the same server Qlane's pre-registered Cursor client. So Cursor signs in as Qlane's verified Cursor client, and an organization's default client policy admits it.

Connected a Qlane server with an earlier version of this plugin? After the update, authenticate the qlane server. It signs in as the verified client.

Cursor can also import this plugin from a Claude Code installation. An imported copy stays at the version Claude Code has installed, and only a version that carries .cursor-plugin/plugin.json signs in as the verified client, so update it there first: claude plugin update qlane@qlane-plugin.

Kiro

In Kiro, choose Import power from GitHub and enter the plugin folder's URL, https://github.com/qlaneai/agent-plugin/tree/main/plugins/qlane. Kiro reads the folder's plugin.json, the portable manifest, and loads the plugin without changes. Then authenticate the qlane server.

Codex

Codex splits this across two places: you register the marketplace on the CLI, but you install from it in the ChatGPT desktop app. There is no CLI install command.

codex plugin marketplace add qlaneai/agent-plugin

Then open the Plugins Directory in the ChatGPT desktop app and install qlane from it. Codex prompts you to sign in the first time it calls the server.

Without the plugin

Any MCP client can point at the endpoint directly. For Claude Code:

claude mcp add --transport http --scope user qlane https://mcp.qlane.ai/mcp

The same URL serves every region.

What it can do

Set up a test target for a repository, read your projects and test targets, and see what Qlane found when it tested your pull requests: each run's verdicts, the defects it saw and the evidence behind them. Three skills put that to work:

  • /qlane:setup-env takes a repository to a test target whose configuration check passed. It finds the Qlane project the repository belongs to, or creates one, prepares the repository from Qlane's published setup pages (docs.qlane.ai/connect), creates the test target, and has you enter its variables and test credentials on a Qlane page — they never pass through the editor. It asks before every change to your repository or to Qlane's configuration.
  • /qlane:bugs shows what Qlane found on a pull request or a recent run — defects, failing and blocked cases, the steps to reproduce them and their evidence — then looks for the likely cause in your repository and proposes a fix. It only reads from Qlane, and it changes your code only when you say yes.
  • /qlane:test-local-changes asks Qlane to plan tests for a local diff before a pull request exists, then exercises them against your running app.

What it sends, and where

The plugin runs no code of its own: it has no hooks and no local servers. Everything below is done by Claude, following the skills, in your session.

  • Qlane's server, mcp.qlane.ai. It checks your sign-in and passes each request on to the region that hosts your organization, mcp-eu.qlane.ai or mcp-us.qlane.ai. Requests to Qlane pass through Cloudflare's global network on their way to your region, which can include locations outside it. Their content is not stored there. The skills send your repository's remote URL to find its Qlane project, and the identifiers of the projects, test targets, runs and pull requests they read. /qlane:setup-env sends a test target's configuration when it creates or changes one, including the repository's name and the branch the test target uses, or, for a URL target, the names of the repositories you link and the branch the deployed site is built from. /qlane:test-local-changes sends your local diff, with a short title and description, for Qlane to plan tests from. Variable values and test-credential passwords never pass through the plugin: you enter them on a Qlane page.
  • GitHub, through your own GitHub CLI. If gh is installed, /qlane:bugs runs gh pr view to find the pull request for your current branch, with your own GitHub sign-in. Without it, the skill asks you for the pull request number.
  • docs.qlane.ai. /qlane:setup-env downloads up to three public setup pages with curl. It sends nothing to them.
  • Your own git remote. /qlane:setup-env commits the files it prepared and pushes them to the branch your test target uses, with your own git access, and only after you say yes.
  • Your local Docker. /qlane:setup-env checks your Compose file with docker compose config and may start your app's main service to confirm it boots. Docker can then pull images from the registries your Compose file names.
  • Your running app. /qlane:test-local-changes drives your locally running app in a browser, at the address you give it, when your session has browser tooling.

Everything else the skills read, such as git log, git diff and your files, stays in your session.

License

MIT.