Hans Aschauer 8b6b237be9 Add test CLI tools for executor testing

- simple_test_cli.py: Working CLI for simple (stateless) backend
  - Supports one-shot execution and interactive REPL
  - Uses PassthroughValidator and wrappers to bypass security for testing
  - Skips resource limits to avoid cgroupv2 issues in rootless Podman
- test_containers.py: Container verification script (all tests passing)
- simple_test_cli_README.md: Documentation for test tools

Note: Jupyter backend has connection file timing issues (future work)

2026-02-07 09:53:49 +01:00

1.9 KiB

Raw Blame History

Simple Test CLI for MCP-Forge

A minimal testing tool for MCP-Forge execution backends.

Features

✅ Working:

Simple (stateless) backend with one-shot and interactive execution
No security restrictions (for testing only)
No resource limits (avoids cgroupv2 issues in rootless Podman)

❌ Not Working:

Jupyter (stateful) backend - has connection file timing issues
MCP tool injection - not supported in this simplified version

Usage

One-Shot Execution (Simple Backend)

# Execute Python code and exit
./simple_test_cli.py --execute 'print(2 + 2)'

# More complex code
./simple_test_cli.py --execute 'x = [1, 2, 3]; print(sum(x))'

Interactive REPL (Simple Backend)

# Start interactive shell
./simple_test_cli.py

# In the shell:
>>> x = 42
>>> print(x * 2)
84
>>> .exit

Special Commands

.exit, .quit - Exit the shell
.help - Show help
.clear - Clear screen

Requirements

Rootless Podman with user socket at $XDG_RUNTIME_DIR/podman/podman.sock
Container images built:
- mcp-forge/python:3.12 - For simple backend
- mcp-forge/jupyter:latest - For Jupyter backend (not working yet)

Limitations

This is a testing tool only:

No security validation
No resource limits
No MCP tool injection
No proper error handling for production use
Jupyter backend has connection file issues

For production use, use the full MCP-Forge server with proper configuration.

Testing Containers

To test if containers work:

# Test simple execution
./test_containers.py

# Should output:
# ✓ Simple Execution: PASS
# ✓ Jupyter Kernel: PASS

Architecture

PassthroughValidator - Dummy validator that allows all operations
SimpleExecutorWrapper - Wraps CodeExecutor, skips resource limits
JupyterKernelWrapper - Wraps JupyterKernelManager (has issues)

The CLI creates temporary directories for logs and connection files, cleans up on exit.

1.9 KiB Raw Blame History