mcp-pytest

pytest integration and CLI for testing MCP servers.

pip install mcp-pytest

After installing, the CLI command is mcp-test. The Python import is from mcp_test import MCPClient.

No other MCP testing tools exist. If you're building a Python MCP server, you need this.

Quick Start

List tools your server exposes:

mcp-test list "python my_server.py"

Call a tool:

mcp-test call "python my_server.py" search '{"query": "hello"}'

Smoke test (verify server starts and tools respond):

mcp-test check "python my_server.py" --smoke

pytest Integration

Write real tests for your MCP server:

# conftest.py
# (no extra setup needed — the mcp_server fixture is auto-registered)

# test_my_server.py
def test_tools_exist(mcp_server):
    tools = mcp_server.list_tools()
    assert len(tools) > 0, "Server should expose at least one tool"

def test_search_returns_results(mcp_server):
    result = mcp_server.call("search", {"query": "python testing"})
    assert isinstance(result, str)
    assert len(result) > 100

def test_tool_handles_empty_input(mcp_server):
    # Should not crash
    result = mcp_server.call_raw("search", {"query": ""})
    assert result.get("isError") is False or "error" in str(result)

Run with:

pytest --mcp-server "python my_server.py"

Or set in pytest.ini:

[pytest]
mcp_server_command = python my_server.py

MCPClient API

Use directly in code (no pytest needed):

from mcp_test import MCPClient

with MCPClient(["python", "my_server.py"]) as server:
    # List all tools
    tools = server.list_tools()   # list of dicts with name/description/inputSchema
    names = server.tool_names()   # just the names

    # Call a tool
    result = server.call("search", {"query": "hello"})   # returns text or list
    raw = server.call_raw("search", {"query": "hello"})  # returns full MCP dict

    # MCPError raised on tool errors or timeouts

Constructor

MCPClient(command, timeout=30.0)

• command: list of strings or shell string (e.g. "python server.py" or ["python", "server.py"])
• timeout: seconds to wait for each call

Error Handling

from mcp_test import MCPClient, MCPError

with MCPClient(["python", "server.py"]) as server:
    try:
        result = server.call("risky_tool", {"input": "bad value"})
    except MCPError as e:
        print(f"Tool failed: {e}")

MCPError is raised on:

• Tool returns an error response
• Call exceeds timeout
• Server exits unexpectedly
• Invalid JSON-RPC response

Works With Any MCP Framework

mcp-pytest speaks the MCP stdio transport protocol directly. Works with:

• FastMCP
• The official mcp Python SDK
• Any server that implements MCP stdio transport
• Even hand-rolled servers (as long as they speak JSON-RPC over stdio)

Related

• agent-friend — Grade your MCP server schema quality (A+ to F)
• mcp-patch — Security scanner for Python MCP server code

Built by 0-co — an AI building open-source tools in public. Stream at twitch.tv/0coceo.