Local UI to run and train LLMs and diffusion models. Supports GGUF, MLX, Qwen3.8, DeepSeek-V4.1, MiniMax-H3, Gemma 4, FLUX and more.
Python agentaichatgptdeepseekfine-tuninggemmaimage-generationllamallmllmsopenaipythonqwenreinforcement-learningself-hostedstable-diffusiontext-to-speechttsuiunsloth
# SPDX-License-Identifier: AGPL-3.0-only
# Copyright 2026-present the Unsloth AI Inc. team. All rights reserved. See /studio/LICENSE.AGPL-3.0
import hashlib
import io
import subprocess
import sys
import tarfile
from concurrent.futures import ThreadPoolExecutor
from pathlib import Path
from unittest.mock import Mock
import pytest
from integrations.blender import runtime
def archive_bytes(extra = None):
buffer = io.BytesIO()
entries = {
"root/mcp/blmcp/__init__.py": b"",
"root/mcp/blmcp/data/prompts.yml": b"initial_instructions: |\n # Bundled Manuals\n old\n # Executing Code\n",
"root/mcp/blmcp/data/api/ignored.rst": b"not installed",
"root/mcp/blmcp/tools/search_api_docs.py": b"not installed",
}
if extra:
entries[extra] = b"unsafe"
with tarfile.open(fileobj = buffer, mode = "w:gz") as archive:
directory = tarfile.TarInfo("root/mcp/blmcp/data")
directory.type = tarfile.DIRTYPE
archive.addfile(directory)
for name, data in entries.items():
member = tarfile.TarInfo(name)
member.size = len(data)
archive.addfile(member, io.BytesIO(data))
return buffer.getvalue()
def test_failed_download_retry_and_concurrent_offline_cache(tmp_path, monkeypatch):
data = archive_bytes()
monkeypatch.setattr(runtime, "cache_root", lambda: tmp_path)
monkeypatch.setattr(runtime, "ARCHIVE_SHA256", hashlib.sha256(data).hexdigest())
download = Mock(side_effect = [b"corrupt", data])
monkeypatch.setattr(runtime, "_download", download)
with pytest.raises(ValueError, match = "checksum"):
runtime.ensure_runtime()
assert not runtime.runtime_path().exists()
assert not list(runtime.runtime_path().parent.glob(".install-*"))
with ThreadPoolExecutor(max_workers = 2) as pool:
paths = list(pool.map(lambda _: runtime.ensure_runtime(), range(2)))
assert paths[0] == paths[1] and download.call_count == 2
assert (paths[0] / ".ready").is_file()
assert not (paths[0] / "blmcp/data/api").exists()
assert not (paths[0] / "blmcp/tools/search_api_docs.py").exists()
assert "Bundled Manuals" not in (paths[0] / "blmcp/data/prompts.yml").read_text()
def test_verified_archive_still_rejects_traversal(tmp_path, monkeypatch):
data = archive_bytes("root/mcp/blmcp/../../escape.py")
monkeypatch.setattr(runtime, "ARCHIVE_SHA256", hashlib.sha256(data).hexdigest())
with pytest.raises(ValueError, match = "Unsafe"):
runtime._extract(data, tmp_path / "runtime")
assert not (tmp_path / "escape.py").exists()
def test_launcher_runs_v1_fastmcp_runtime_on_installed_mcp(tmp_path):
package = tmp_path / "blmcp"
package.mkdir()
(tmp_path / ".ready").write_text("x")
(package / "__init__.py").write_text(
"import asyncio\n"
"from mcp.server.fastmcp import FastMCP, Image\n"
"def main():\n"
" server = FastMCP('blender-mcp', instructions='x')\n"
" @server.tool()\n"
" def fail() -> int:\n"
" raise RuntimeError('Blender error: NameError')\n"
" try:\n"
" asyncio.run(server.call_tool('fail', {}))\n"
" except Exception as exc:\n"
" print(exc)\n"
" print(Image(data=b'x', format='png').data)\n"
" return 0\n"
)
launcher = Path(runtime.__file__).with_name("launcher.py")
result = subprocess.run(
[sys.executable, str(launcher), str(tmp_path)], capture_output = True, text = True
)
assert result.returncode == 0, result.stderr
assert result.stdout.splitlines() == [
"Error executing tool fail: Blender error: NameError",
"b'x'",
]