Blog

How to write and run tests from Claude with MCP

You can write and run tests from an AI assistant when your test platform speaks the Model Context Protocol (MCP). With SmartestQA you connect Claude, Claude Code, Cursor or another MCP client, and the assistant works on your real test project: it plans, writes and runs tests, and it changes nothing until you approve.

The problem

An AI assistant can draft a test from a description. On its own it cannot know which requirements already have a test, or why last night's run went red. You end up copying context into the chat and copying code back out, and nobody can tell afterwards what the assistant changed.

MCP solves the access part: it lets an assistant call the tools of another system. Access alone is not enough, though. A QA lead also needs to know that the assistant cannot quietly rewrite a regression suite.

How SmartestQA does it

The assistant works inside the platform, under your own permissions, and every change passes through you.

  • It sees the real project. Requirements, tests and results, not a description of them.
  • It proposes, you approve. Each change arrives as a diff. You say yes, ask for a change, or say no.
  • It runs what it writes. A new test is tried at once, so you approve something that already passes.
  • It leaves a trail. Every change made through the assistant is in the audit log.

Questions and answers

Which AI assistants work with SmartestQA?

Claude on claude.ai, Claude Desktop, Claude Code, Cursor, and any other MCP client that can start a local server.

Can the assistant change my tests without asking?

No. Every change is proposed first and written only after you approve it.

Does SmartestQA see my conversation with the assistant?

No. It receives only the tool calls the assistant makes. The conversation stays with your AI provider.

Where to read more

See MCP and AI assistants, or the setup in the documentation: MCP & AI assistants.