Testing AI-generated code during vibe coding

An AI coding agent can write a change quickly, but it still needs a trustworthy test result. TimmyTest runs the repository's existing suite locally, reports likely missing tests, and produces concise context for the next debugging step.

Start with the existing tests

Install the tagged release in a Python 3.11+ environment, open the project you want to test, and run one audit:

python -m pip install "git+https://github.com/tugrakaymakcioglu/TimmyTest.git@v2.0.1"
timmytest check . --no-copy-prompt

The audit detects the test setup, runs the suite and summarizes failures and likely source-to-test gaps. TimmyTest does not create a new test framework or prove that every behavior is covered. Review its gap suggestions against the repository's actual tests.

Let the agent choose the right tool

Agent task	TimmyTest choice
Tests fail after a code change	`timmytest run . --only-failures` or MCP tool `timmytest_run`
Find likely missing tests without executing code	`timmytest scan .` or `timmytest_scan`
Get a full report before debugging	`timmytest check . --no-copy-prompt` or `timmytest_check`
Need a compact handoff for another agent	`timmytest agent .` or `timmytest_prompt`

For automatic selection within a repository, run timmytest integrate. It writes agent instructions for Claude Code, Codex, Cursor and Copilot and adds a Cursor MCP configuration. Review those generated files before committing them. An agent cannot discover or call TimmyTest unless its environment has the CLI or MCP server installed and exposed.

Configure the MCP server

In an MCP client that accepts a local stdio server, register the installed executable:

{
  "mcpServers": {
    "timmytest": { "command": "timmytest", "args": ["mcp"] }
  }
}

Client configuration formats differ. Use the client's current MCP setup instructions if it does not accept this JSON format. The MCP tool descriptions tell the agent which tool fits a test run, failure diagnosis or static gap scan; the agent and its configuration still decide whether to call it.

What the result means

The local scan and test run make no AI API calls. Sending a handoff prompt to an AI agent does consume that agent's tokens. A rule-based diagnosis is a lead, not proof of the root cause. If TimmyTest cannot run your suite or a command fails, use your project's standard test runner and report the unsupported setup.

See the terminal demo, 2.0.1 release and full command reference.