mcp-assert 0.12.3

Deterministic correctness testing for MCP servers

mcp-assert

The testing standard for MCP servers. Lint, test, and fuzz. No SDK required. No LLM required.

This is the PyPI distribution of mcp-assert, a single Go binary that connects to your MCP server over stdio, SSE, or HTTP, runs 24 static analysis rules, calls tools with real arguments, and asserts results against YAML expectations.

Install

pip install mcp-assert

Usage

# Zero-config audit (finds crashes)
mcp-assert audit --server "python my_server.py"

# Static analysis (finds schema issues)
mcp-assert lint --server "python my_server.py"

# Auto-fix schema issues
mcp-assert lint --server "python my_server.py" --fix

# Adversarial fuzzing
mcp-assert fuzz --server "python my_server.py"

# Run YAML assertions
mcp-assert run --suite evals/

# CI with threshold
mcp-assert ci --suite evals/ --threshold 95

Or use as a Python module:

python -m mcp_assert lint --server "python my_server.py"

pytest Integration

pip install pytest-mcp-assert
pytest --mcp-suite evals/

Each YAML file becomes a pytest Item. Configure via pyproject.toml:

[tool.pytest.ini_options]
mcp_suite = "evals/"
mcp_fixture = "fixtures/"

Lint (24 Rules)

mcp-assert lint --server "python my_server.py"

  E  E103   create_user    Required parameter "email" has no description
  W  W109   search         User-facing parameter "query" has no examples
  W  W108   delete_user    Tool name implies mutation but description doesn't acknowledge

# Auto-generate fixes
mcp-assert lint --server "python my_server.py" --fix

# Strict mode for CI
mcp-assert lint --server "python my_server.py" --strict

Links

• Documentation
• GitHub
• Error Reference (24 error codes)
• Scorecard (102 servers scanned, 4,794 issues found)
• pytest plugin

License

MIT