mcp-pytest-runner 0.2.0

MCP server providing opinionated pytest execution interface for AI agents

mcp-pytest-runner

MCP server providing opinionated pytest execution interface for AI agents.

Overview

mcp-pytest-runner is a Model Context Protocol (MCP) server that enables AI coding assistants to execute pytest with intelligent test selection, structured result interpretation, and context-aware recommendations.

Status

This project is in active development. MCP server implementation complete with stdio transport, test discovery, and test execution capabilities.

Installation

Install via uvx for immediate use:

uvx mcp-pytest-runner

Or add to your Python environment:

pip install mcp-pytest-runner

MCP Integration

mcp-pytest-runner provides a Model Context Protocol server that enables AI coding assistants to execute pytest with intelligent test selection and structured result interpretation.

Claude Code Configuration

Add mcp-pytest-runner to your Claude Code MCP settings using the claude mcp add command:

claude mcp add pytest uvx mcp-pytest-runner

Or manually configure by editing your MCP settings. The configuration file location depends on your platform:

• macOS: ~/Library/Application Support/Claude/claude_desktop_config.json
• Linux: ~/.config/Claude/claude_desktop_config.json
• Windows: %APPDATA%\Claude\claude_desktop_config.json

Example manual configuration:

{
  "mcpServers": {
    "pytest": {
      "command": "uvx",
      "args": ["mcp-pytest-runner"]
    }
  }
}

Why uvx? Ensures you always use the latest version without manual updates. Alternative: use mcp-pytest-runner directly if installed in your system Python.

Available Tools

discover_tests

Discover pytest test structure without executing tests.

Parameters:

• path (optional): Directory or file path to discover tests within (default: project root)
• pattern (optional): Test file pattern (default: test_*.py or *_test.py)

Returns: Hierarchical test organization with node IDs for subsequent execution.

Why use this? Understand test suite structure before execution. Enables intelligent test selection in TDD workflows.

execute_tests

Execute pytest tests with validated parameters.

Parameters:

• node_ids (optional): Specific test node IDs to execute (e.g., ["tests/test_user.py::test_login"])
• markers (optional): Pytest marker expression (e.g., "not slow and integration")
• keywords (optional): Keyword expression for test name matching
• verbosity (optional): Output verbosity level (-2 to 2)
• failfast (optional): Stop execution on first failure
• maxfail (optional): Stop after N failures
• show_capture (optional): Include captured stdout/stderr
• timeout (optional): Execution timeout in seconds

Returns: Structured results including pass/fail status, error messages, stack traces, and test summary.

Exit Code Handling:

• Success response (exit codes 0, 1, 5): Structured test results with failure details
• Error response (exit codes 2, 3, 4): pytest configuration or execution errors

Why this design? Test failures are normal TDD workflow outcomes. The tool succeeds when pytest executes successfully, regardless of test pass/fail status.

Connection Testing

Verify the MCP server is working correctly:

1. Restart Claude Code after updating configuration
2. Check MCP connection: Look for pytest tools in Claude's available tools
3. Test discovery: Ask Claude to "discover tests in this project"
4. Test execution: Ask Claude to "run all tests"

Troubleshooting

MCP server not appearing in Claude Code:

• Verify JSON configuration syntax (no trailing commas)
• Check file path matches your platform
• Restart Claude Code completely (quit and relaunch)

Tests not discovered:

• Verify pytest is installed in your project environment
• Check path parameter points to valid test directory
• Ensure pattern matches your test file naming convention

Test execution failures:

• Review error response for pytest configuration issues
• Verify node IDs from discovery match execution parameters
• Check timeout parameter if tests run longer than default

Development

This project uses Nix for reproducible development environments. To get started:

# Enter development shell
nix develop

# Run tests
pytest

# Run type checking
mypy src

# Run linting
ruff check src tests

# Run security scanning
bandit -r src

License

MIT License - See LICENSE file for details.