MCP Server

The Model Context Protocol (MCP) is an open standard for giving AI assistants access to external systems. QA Sphere serves an MCP server directly, so an assistant can read and work with your test cases, runs, and results instead of you copying data into a chat window.

Typical uses:

• "Which test cases cover the checkout flow?" answered from your real library, inside your editor
• Referencing a specific QA Sphere test case while writing the automated test for it
• Summarizing what failed in the latest run without leaving your IDE
• Drafting new test cases from a spec and creating them in QA Sphere

MCP access is included on every plan, and nothing needs to be installed locally.

Setting It Up

Open Settings → MCP Server in QA Sphere. The page asks you to pick your assistant and an access level, walks through adding an API key, and then shows the exact setup instructions and configuration for that client:

• Claude Code
• Codex
• Gemini CLI
• VS Code
• Cursor

Copy the configuration it generates and paste it where your client keeps its MCP settings. Because the configuration embeds your workspace URL and API key, take it from this page rather than assembling it by hand.

The hosted server stays current with the QA Sphere API and exposes the full set of tools allowed by your role — 27 in total.

Access Levels

Two access levels are available, chosen when you set the server up:

• Standard — read and write. The assistant can create and update test cases, runs, and results.
• Read-Only — the assistant can query and summarize, but cannot change anything in your workspace.

Start with Read-Only. It covers the most common uses (search, summarize, explain coverage) and removes any risk of an assistant mutating your library while you are still building trust in it. Move to Standard when you actually want the assistant to write back.

Your role also constrains what the server can reach: the tools exposed are those your account is permitted to use. See Users and Permissions.

Keeping It Safe

A few habits worth adopting:

• Scope the key to the least you need. An API key inherits the role of the user who created it, and the public API is per-user rate limited at 20 requests per second. A Read-Only setup or a lower-privileged user account limits the blast radius of a confused assistant.
• Use a dedicated key. A key created specifically for MCP can be revoked without disrupting your CI pipelines or the CLI.
• Review writes. With Standard access an assistant can create and modify test cases. Read what it proposes before accepting, the same as you would with generated code.

Retired: the Standalone qasphere-mcp Package

Before QA Sphere served MCP directly, a standalone server was published as the qasphere-mcp npm package and run locally through npx.

If you still have it configured, migrate: remove the qasphere-mcp entry from your client's MCP configuration, then follow the setup instructions on the Settings → MCP Server page. The hosted server needs no local install, so there is nothing left to uninstall beyond that configuration block. While the standalone server is running it prints a migration notice once per session.

MCP, the CLI, and the API

Three ways to reach the same data, suited to different jobs:

If you want an AI coding agent to drive QA Sphere through the CLI rather than over MCP, the CLI ships a skill for exactly that — see Agent Skill.

Troubleshooting

Still stuck? Contact us at sorted@qasphere.com.