Installs Qlane MCP into your AI coding editor, so your agent can reach your Qlane projects, test targets and test-case counts without leaving the editor.
Qlane runs two independent regional stacks. The plugin registers one server for each, and you connect the one hosting your organization:
| Server | Endpoint |
|---|---|
qlane-eu |
https://mcp-eu.qlane.ai/api/mcp |
qlane-us |
https://mcp-us.qlane.ai/api/mcp |
Authenticate that one and leave the other alone. An unconnected server contributes no tools and costs nothing; it is not an error state. If you connect the wrong one, every call returns a 403 naming the correct host — nothing is lost, and the fix is to connect the other server.
Both are plain static URLs, so this works identically on every editor. Nothing is configured, substituted, or typed.
Qlane MCP requires a Qlane account. Your editor prompts you to sign in the first time it calls the server.
claude plugin marketplace add qlaneai/agent-plugin
claude plugin install qlane@qlane-pluginThen run /mcp and authenticate qlane-eu or qlane-us — whichever hosts your
organization. There is no configuration step and nothing to type.
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 reads the root plugin.json and loads this plugin without changes. Add the repository
as a marketplace, install qlane, then authenticate the server for your region.
Same as Cursor — add the repository, install, authenticate your region's server.
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-pluginThen 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.
You will see qlane-eu and qlane-us listed as Pending approval. That is expected and
harmless: Claude Code reads a .mcp.json at a repository root as project-scoped MCP
configuration, so this repo's .mcp.json — the plugin's own manifest — is being read as if it
were yours. Decline it, and install the plugin with the commands above instead.
Any MCP client can point at the endpoint directly. For Claude Code:
claude mcp add --transport http --scope user qlane https://mcp-eu.qlane.ai/api/mcpSwap the host for mcp-us.qlane.ai if your organization is US-hosted.
Read your projects, test targets and test-case counts, and plan tests for a local diff before a pull request exists. Planning a test consumes Qlane credits.
npm ci
npm run checkcheck runs Prettier, markdownlint, JSON Schema validation of the portable manifests, and the
unit tests. It makes no network calls, so it works offline and in a sandboxed CI runner.
npm run format # apply Prettier
npm run lint:md:fix # apply markdownlint's safe fixesTwo checks need the Claude Code CLI and so are not part of check, which has to run on a
runner that does not have it:
claude plugin validate --strict . # manifest shapes; --strict makes warnings fatal
claude plugin tag --dry-run # what the next release tag would beThe version lives in plugin.json only. The marketplace entry deliberately declares no
version of its own: plugin.json wins at install time, so a second copy could only ever
agree or be silently ignored.
No manifest may contain a placeholder. Both regional URLs are literal strings in every
manifest. ${user_config.*} resolves only in the Claude Code CLI — the portable Agent Plugins
schema cannot even express userConfig (additionalProperties: false), and claude.ai treats
an MCP url as an opaque connector identity, so a templated host reaches registration verbatim
and fails to resolve. A test enforces this.
MIT.