Data Science — the complete, unedited output of the deterministic mcptrustchecker engine v1.13.0, scanned . Every finding, capability tag and score component below is exactly what the engine produced — no AI, no post-processing.
{
"tool": {
"name": "mcptrustchecker",
"version": "1.13.0",
"methodologyVersion": "mcptrustchecker-1.13"
},
"target": {
"id": "data-science-mcp",
"source": {
"kind": "package",
"origin": "data-science-mcp"
},
"server": {
"name": "data-science-mcp"
}
},
"grade": "A",
"score": {
"score": 92,
"threatScore": 100,
"grade": "A",
"band": "A",
"categorySubtotals": {
"injection": 0,
"exfiltration": 0,
"permissions": 0,
"supply-chain": 0,
"network": 0,
"hygiene": 0
},
"vector": [
{
"kind": "client",
"term": "capability-exposure",
"level": "high",
"label": "capability blast radius (high) — client exposure if the model is manipulated",
"appliedPenalty": 6
},
{
"kind": "client",
"term": "verification-discount",
"level": "none",
"label": "publisher verification (unlinked) — no provenance/repo link, but the shipped source was fully read",
"appliedPenalty": 2
},
{
"kind": "client",
"term": "coverage-honesty",
"level": "source",
"label": "inspection depth (source) — how much of the target the scan could see",
"appliedPenalty": 0
}
],
"gatesFired": [],
"methodologyVersion": "mcptrustchecker-1.13"
},
"capability": {
"level": "high",
"reasons": [
"can execute shell commands or code"
],
"tags": [
"code-exec"
]
},
"coverage": {
"level": "source",
"inputs": {
"toolSurface": true,
"implementationSource": true,
"packageMetadata": true,
"liveTransport": false
},
"caveats": [
"Tools were statically extracted from the published source (38 recovered), not enumerated from a running server. Tool-poisoning, Unicode-smuggling, capability and toxic-flow analysis ran on this inferred surface, but a mis-parsed registration could be missed or mis-attributed, so tool-derived findings are capped below “confirmed”. To grade the real runtime surface, scan the running server: --command \"npx -y <package>\"."
]
},
"findings": [
{
"ruleId": "MTC-SRC-001",
"title": "Dynamic code execution in server code (data_science_mcp/inference/base.py)",
"category": "permissions",
"severity": "high",
"confidence": "strong",
"description": "In the server's implementation (`data_science_mcp/inference/base.py:6`): Evaluating strings as code is the most direct RCE primitive; if any tool input reaches it, the server executes attacker-chosen code. This is read from the code itself — not from the tool description — so a poisoned server cannot hide it behind honest-looking metadata.",
"remediation": "Review this call path: confirm it never receives unsanitized tool input, constrain it, or remove it. Treat a server whose code reaches these sinks as high-capability regardless of what its tools claim.",
"location": {
"kind": "server",
"name": "data_science_mcp/inference/base.py"
},
"evidence": "post-train reliability eval (one completion per case). Both vLLM and SGLang expose the **same OpenAI-compatible HTTP pr",
"owasp": "LLM05:2025 Improper Output Handling",
"data": {
"rule": "MTC-SRC-001",
"file": "data_science_mcp/inference/base.py",
"line": 6,
"nonRuntime": false
}
},
{
"ruleId": "MTC-SRC-001",
"title": "Dynamic code execution in server code (data_science_mcp/kernels/_runner.py)",
"category": "permissions",
"severity": "high",
"confidence": "strong",
"description": "In the server's implementation (`data_science_mcp/kernels/_runner.py:49`): Evaluating strings as code is the most direct RCE primitive; if any tool input reaches it, the server executes attacker-chosen code. This is read from the code itself — not from the tool description — so a poisoned server cannot hide it behind honest-looking metadata.",
"remediation": "Review this call path: confirm it never receives unsanitized tool input, constrain it, or remove it. Treat a server whose code reaches these sinks as high-capability regardless of what its tools claim.",
"location": {
"kind": "server",
"name": "data_science_mcp/kernels/_runner.py"
},
"evidence": "] = {} try: exec(compile(src, \"<candidate>\", \"exec\"), namespace) # noqa: S102 — sandboxed subprocess ex",
"owasp": "LLM05:2025 Improper Output Handling",
"data": {
"rule": "MTC-SRC-001",
"file": "data_science_mcp/kernels/_runner.py",
"line": 49,
"nonRuntime": false
}
},
{
"ruleId": "MTC-SRC-001",
"title": "Dynamic code execution in server code (data_science_mcp/training_pipeline.py)",
"category": "permissions",
"severity": "high",
"confidence": "strong",
"description": "In the server's implementation (`data_science_mcp/training_pipeline.py:238`): Evaluating strings as code is the most direct RCE primitive; if any tool input reaches it, the server executes attacker-chosen code. This is read from the code itself — not from the tool description — so a poisoned server cannot hide it behind honest-looking metadata.",
"remediation": "Review this call path: confirm it never receives unsanitized tool input, constrain it, or remove it. Treat a server whose code reaches these sinks as high-capability regardless of what its tools claim.",
"location": {
"kind": "server",
"name": "data_science_mcp/training_pipeline.py"
},
"evidence": "fn for the reliability eval (defaults to a no-op echo when omitted so the pipeline still completes on CPU).",
"owasp": "LLM05:2025 Improper Output Handling",
"data": {
"rule": "MTC-SRC-001",
"file": "data_science_mcp/training_pipeline.py",
"line": 238,
"nonRuntime": false
}
},
{
"ruleId": "MTC-SRC-002",
"title": "Shell/command execution in server code (data_science_mcp/kernels/_runner.py)",
"category": "permissions",
"severity": "high",
"confidence": "strong",
"description": "In the server's implementation (`data_science_mcp/kernels/_runner.py:49`): Spawning a shell/process is command-execution capability; with unsanitized tool input it is command injection / RCE. This is read from the code itself — not from the tool description — so a poisoned server cannot hide it behind honest-looking metadata.",
"remediation": "Review this call path: confirm it never receives unsanitized tool input, constrain it, or remove it. Treat a server whose code reaches these sinks as high-capability regardless of what its tools claim.",
"location": {
"kind": "server",
"name": "data_science_mcp/kernels/_runner.py"
},
"evidence": "] = {} try: exec(compile(src, \"<candidate>\", \"exec\"), namespace) # noqa: S102 — sandboxed subprocess ex",
"owasp": "LLM05:2025 Improper Output Handling",
"data": {
"rule": "MTC-SRC-002",
"file": "data_science_mcp/kernels/_runner.py",
"line": 49,
"nonRuntime": false
}
},
{
"ruleId": "MTC-SRC-002",
"title": "Shell/command execution in server code (data_science_mcp/kernels/kernel_verifier.py)",
"category": "permissions",
"severity": "high",
"confidence": "strong",
"description": "In the server's implementation (`data_science_mcp/kernels/kernel_verifier.py:53`): Spawning a shell/process is command-execution capability; with unsanitized tool input it is command injection / RCE. This is read from the code itself — not from the tool description — so a poisoned server cannot hide it behind honest-looking metadata.",
"remediation": "Review this call path: confirm it never receives unsanitized tool input, constrain it, or remove it. Treat a server whose code reaches these sinks as high-capability regardless of what its tools claim.",
"location": {
"kind": "server",
"name": "data_science_mcp/kernels/kernel_verifier.py"
},
"evidence": "proc = subprocess.run( [self.python_exe, \"-m\", \"data_science_mcp.kernels._runner\",",
"owasp": "LLM05:2025 Improper Output Handling",
"data": {
"rule": "MTC-SRC-002",
"file": "data_science_mcp/kernels/kernel_verifier.py",
"line": 53,
"nonRuntime": false
}
},
{
"ruleId": "MTC-SRC-001",
"title": "Dynamic code execution in packaging/dev tooling (tests/test_launch.py)",
"category": "permissions",
"severity": "low",
"confidence": "heuristic",
"description": "In a packaging/dev/install script (shipped, but not the server runtime) (`tests/test_launch.py:2`): Evaluating strings as code is the most direct RCE primitive; if any tool input reaches it, the server executes attacker-chosen code. This is read from the code itself — not from the tool description — so a poisoned server cannot hide it behind honest-looking metadata.",
"remediation": "Review this call path: confirm it never receives unsanitized tool input, constrain it, or remove it. Treat a server whose code reaches these sinks as high-capability regardless of what its tools claim.",
"location": {
"kind": "server",
"name": "tests/test_launch.py"
},
"evidence": "ed launcher + benchmark eval (CONCEPT:DS-AHE.trainer.concept-4/006). Config builders and the ``accelerate launch`` argv",
"owasp": "LLM05:2025 Improper Output Handling",
"data": {
"rule": "MTC-SRC-001",
"file": "tests/test_launch.py",
"line": 2,
"nonRuntime": true
}
},
{
"ruleId": "MTC-SRC-002",
"title": "Shell/command execution in packaging/dev tooling (scripts/security_sanitizer.py)",
"category": "permissions",
"severity": "low",
"confidence": "heuristic",
"description": "In a packaging/dev/install script (shipped, but not the server runtime) (`scripts/security_sanitizer.py:136`): Spawning a shell/process is command-execution capability; with unsanitized tool input it is command injection / RCE. This is read from the code itself — not from the tool description — so a poisoned server cannot hide it behind honest-looking metadata.",
"remediation": "Review this call path: confirm it never receives unsanitized tool input, constrain it, or remove it. Treat a server whose code reaches these sinks as high-capability regardless of what its tools claim.",
"location": {
"kind": "server",
"name": "scripts/security_sanitizer.py"
},
"evidence": "try: result = subprocess.run( [\"git\", \"ls-files\", \"--cached\", \"--others\", \"--exclude-standard\"],",
"owasp": "LLM05:2025 Improper Output Handling",
"data": {
"rule": "MTC-SRC-002",
"file": "scripts/security_sanitizer.py",
"line": 136,
"nonRuntime": true
}
},
{
"ruleId": "MTC-SRC-002",
"title": "Shell/command execution in packaging/dev tooling (tests/conftest.py)",
"category": "permissions",
"severity": "low",
"confidence": "heuristic",
"description": "In a packaging/dev/install script (shipped, but not the server runtime) (`tests/conftest.py:81`): Spawning a shell/process is command-execution capability; with unsanitized tool input it is command injection / RCE. This is read from the code itself — not from the tool description — so a poisoned server cannot hide it behind honest-looking metadata.",
"remediation": "Review this call path: confirm it never receives unsanitized tool input, constrain it, or remove it. Treat a server whose code reaches these sinks as high-capability regardless of what its tools claim.",
"location": {
"kind": "server",
"name": "tests/conftest.py"
},
"evidence": "ngine.sock\") proc = subprocess.Popen( [binary, \"--socket-path\", sock], stdout=subprocess.PIPE,",
"owasp": "LLM05:2025 Improper Output Handling",
"data": {
"rule": "MTC-SRC-002",
"file": "tests/conftest.py",
"line": 81,
"nonRuntime": true
}
},
{
"ruleId": "MTC-SUP-011",
"title": "Package declares no source repository",
"category": "supply-chain",
"severity": "info",
"confidence": "strong",
"description": "\"data-science-mcp\" declares no repository URL, so its published artifact cannot be compared against reviewable source.",
"remediation": "Prefer packages that link to public, reviewable source.",
"location": {
"kind": "package",
"name": "data-science-mcp"
}
}
],
"toxicFlows": [],
"capabilities": [
{
"tool": "dedup_corpus",
"tags": [],
"reasons": {}
},
{
"tool": "decontaminate_corpus",
"tags": [],
"reasons": {}
},
{
"tool": "curate_corpus",
"tags": [],
"reasons": {}
},
{
"tool": "prepare_pretrain_data",
"tags": [],
"reasons": {}
},
{
"tool": "dataset_lineage",
"tags": [],
"reasons": {}
},
{
"tool": "load_dataset",
"tags": [],
"reasons": {}
},
{
"tool": "describe_dataset",
"tags": [],
"reasons": {}
},
{
"tool": "split_dataset",
"tags": [],
"reasons": {}
},
{
"tool": "deep_train_predict",
"tags": [],
"reasons": {}
},
{
"tool": "generate_interpretability_tests",
"tags": [],
"reasons": {}
},
{
"tool": "grade_response",
"tags": [],
"reasons": {}
},
{
"tool": "run_interpretability_suite",
"tags": [],
"reasons": {}
},
{
"tool": "ds_specialize_kernel",
"tags": [],
"reasons": {}
},
{
"tool": "evolve_model_class",
"tags": [],
"reasons": {}
},
{
"tool": "rank_models",
"tags": [],
"reasons": {}
},
{
"tool": "get_pareto_frontier",
"tags": [],
"reasons": {}
},
{
"tool": "fit_model",
"tags": [],
"reasons": {}
},
{
"tool": "predict",
"tags": [],
"reasons": {}
},
{
"tool": "evaluate_model",
"tags": [],
"reasons": {}
},
{
"tool": "cross_validate",
"tags": [],
"reasons": {}
},
{
"tool": "quant_market_making",
"tags": [],
"reasons": {}
},
{
"tool": "quant_microstructure",
"tags": [],
"reasons": {}
},
{
"tool": "quant_sizing",
"tags": [],
"reasons": {}
},
{
"tool": "quant_validation",
"tags": [],
"reasons": {}
},
{
"tool": "quant_statespace",
"tags": [],
"reasons": {}
},
{
"tool": "quant_signals",
"tags": [],
"reasons": {}
},
{
"tool": "quant_forensic",
"tags": [],
"reasons": {}
},
{
"tool": "quant_derivatives",
"tags": [],
"reasons": {}
},
{
"tool": "train_sft",
"tags": [],
"reasons": {}
},
{
"tool": "train_dpo",
"tags": [],
"reasons": {}
},
{
"tool": "train_grpo",
"tags": [],
"reasons": {}
},
{
"tool": "train_reward",
"tags": [],
"reasons": {}
},
{
"tool": "train_ppo",
"tags": [],
"reasons": {}
},
{
"tool": "merge_adapters_ties",
"tags": [],
"reasons": {}
},
{
"tool": "pretrain_model",
"tags": [],
"reasons": {}
},
{
"tool": "train_tokenizer",
"tags": [],
"reasons": {}
},
{
"tool": "build_training_dataset",
"tags": [],
"reasons": {}
},
{
"tool": "compose_reward",
"tags": [],
"reasons": {}
}
],
"surfaceDigest": "d9fc9f07ed102c160614333a4a4420a2cc0fbbf9cd220e1ed8f68df9e36e355b",
"stats": {
"tools": 38,
"prompts": 0,
"resources": 0,
"findingsBySeverity": {
"critical": 0,
"high": 5,
"medium": 0,
"low": 3,
"info": 1
}
}
}