com.aiinfradecoded/mcp-anvil

What we checked

Six static checks, weighted by risk. Every result reflects only what could be observed in the published package and repository — never intent.

• Injection surface

  unscannable25% of grade

  Tool descriptions/manifest scanned for instruction-injection patterns (imperatives at the model, hidden text, 'ignore previous', data-exfil URLs).

  No tool descriptions, server instructions, prompts, or resources were fetched; nothing to scan for injection.

• Supply chain

  warn25% of grade

  Package provenance: namespace verification, repo linkage, maintainer count, account age, postinstall scripts, typosquat distance.

  no resolvable repository linked to the package (provenance unverifiable)

• Credential hygiene

  warn15% of grade

  How the server takes secrets (env vs plaintext config vs hardcoded); secrets appearing in tool schemas.

  tests/test_audit.py: contains a hardcoded OpenAI API key (sk-a…AA (len 33)) — in an example/sample file; verify it is not a real key. tests/test_builtin_tools_expanded.py: contains a hardcoded GitHub personal access token (classic) (ghp_…aa (len 40)) — in an example/sample file; verify it is not a real key. tests/test_stress.py: contains a hardcoded GitHub personal access token (classic) (ghp_…aa (len 40)) — in an example/sample file; verify it is not a real key. Reads secrets from the environment (e.g. examples/custom_tools/_generic_http_tool_template.py) — the recommended intake shape.

• Permission scope

  fail15% of grade

  Declared tools vs. breadth (filesystem, network, exec); flags shell-exec and unbounded filesystem access.

  HIGH: src/mcp_anvil/daemon/discovery.py: references a sensitive credential path (/.cursor/) — the tool-poisoning exfiltration target. HIGH: tests/test_security_round3.py: references a sensitive credential path (/.ssh/) — the tool-poisoning exfiltration target. HIGH: shell/exec capability AND network egress together — download-and-execute risk (scripts/e2e_test.py: subprocess.* call; examples/custom_tools/_generic_http_tool_template.py: outbound HTTP/socket call). HIGH: shell/exec capability AND destructive filesystem writes together (scripts/e2e_test.py: subprocess.* call; src/mcp_anvil/commands/keys_cmd.py: destructive filesystem call (shutil.rmtree/os.remove/etc.)). discloses shell/exec capability: scripts/e2e_test.py: subprocess.* call discloses shell/exec capability: scripts/send_license_email.py: subprocess.* call discloses shell/exec capability: src/mcp_anvil/audit/http_sandbox.py: subprocess.* call discloses shell/exec capability: src/mcp_anvil/audit/sandbox.py: subprocess.* call discloses shell/exec capability: src/mcp_anvil/builtin_tools/tools_dev.py: subprocess.* call discloses shell/exec capability: src/mcp_anvil/commands/tool_cmd.py: subprocess.* call discloses shell/exec capability: src/mcp_anvil/commands/tray_cmd.py: subprocess.* call discloses shell/exec capability: src/mcp_anvil/daemon/process.py: subprocess.* call discloses shell/exec capability: tests/test_builtin_tools.py: subprocess.* call discloses shell/exec capability: tests/test_runtime.py: subprocess.* call discloses dynamic code execution: src/mcp_anvil/custom_tools/loader.py: uses eval/exec/__import__ discloses dynamic code execution: src/mcp_anvil/daemon/runtime.py: uses eval/exec/__import__ discloses dynamic code execution: src/mcp_anvil/daemon/api/app.py: uses eval/exec/__import__ discloses dynamic code execution: tests/test_c4_stronger.py: uses eval/exec/__import__ discloses broad/destructive filesystem access: src/mcp_anvil/commands/keys_cmd.py: destructive filesystem call (shutil.rmtree/os.remove/etc.) discloses broad/destructive filesystem access: src/mcp_anvil/commands/reset_cmd.py: destructive filesystem call (shutil.rmtree/os.remove/etc.) discloses broad/destructive filesystem access: src/mcp_anvil/daemon/config.py: destructive filesystem call (shutil.rmtree/os.remove/etc.) discloses broad/destructive filesystem access: tests/test_c4_stronger.py: destructive filesystem call (shutil.rmtree/os.remove/etc.) discloses broad/destructive filesystem access: tests/test_prompts.py: path traversal ('../') combined with file access discloses broad/destructive filesystem access: tests/test_runtime.py: path traversal ('../') combined with file access discloses broad/destructive filesystem access: tests/test_stress.py: open(…, 'w'/'a'/'x') write-mode file access discloses broad/destructive filesystem access: tests/test_stress.py: path traversal ('../') combined with file access discloses broad/destructive filesystem access: tests/test_templates.py: path traversal ('../') combined with file access Capability DISCLOSURE, not a verdict: static analysis sees the primitive is present and reachable, not whether its use is attacker-controlled.

• Version behavior

  unscannable10% of grade

  Diff of tool definitions between versions; new permissions or changed descriptions in a patch release (the postmark-mcp class).

  no prior version to diff (first sight of com.aiinfradecoded/[email protected]); version-drift / rug-pull cannot be evaluated. Cold-start risk is covered by point-in-time checks (injection_surface, supply_chain, credential_hygiene), not here.

• Transport config

  warn10% of grade

  Remote servers: TLS and auth mode (none/token/OAuth). Local servers: whether the manifest indicates it phones home.

  manifest/source references heartbeat behavior — disclosure