com.local-mcp/local-mcp

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

  fail25% of grade

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

  tool "create_draft" param name "bcc": names an outbound copy/exfil destination (BCC/exfil verb). tool "create_draft" param "bcc" description: names an outbound copy/exfil destination (BCC/exfil verb). tool "run_diagnostics" description: names an outbound copy/exfil destination (BCC/exfil verb). tool "run_diagnostics" param "submit" description: names an outbound copy/exfil destination (BCC/exfil verb). tool "send_email" param name "bcc": names an outbound copy/exfil destination (BCC/exfil verb). tool "send_email" param "bcc" description: names an outbound copy/exfil destination (BCC/exfil verb). tool "list_emails" description: bare 'IMPORTANT:' prose in a description (tag form is high-confidence; prose is warn). tool "read_email" description: cross-tool redirection ('instead of call…'). tool "search_emails" description: bare 'IMPORTANT:' prose in a description (tag form is high-confidence; prose is warn). tool "ui_type" description: imperative telling the model it must read/send/attach something. tool "update_reminder" description: bare 'IMPORTANT:' prose in a description (tag form is high-confidence; prose is warn). tool "web_login" param "url" description: off-vendor URL in a description (www.linkedin.com); capability text should not name where data goes.

• Supply chain

  warn25% of grade

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

  install hook "postinstall" present and not a recognized benign build command

• Credential hygiene

  fail15% of grade

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

  tool "recipe_import" has a secret-shaped input parameter "skill_token" — secrets must never be tool parameters (the model would supply/handle the credential). Reads secrets from the environment (e.g. download.js) — the recommended intake shape.

• Permission scope

  fail15% of grade

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

  HIGH: setup.js reads a credential path, executes commands, and makes network calls in the same file — the combination that makes exfiltration possible. references credential paths: setup.js: references a credential path (/.cursor/) has both shell/exec and network egress — capable of download-and-run, though no such path was observed (download.js: imports child_process and calls exec/spawn; setup.js: outbound HTTP/socket call). has both shell/exec and destructive filesystem writes (download.js: imports child_process and calls exec/spawn; download.js: destructive/writing fs.* call (write/unlink/rm/chmod/rename)). discloses shell/exec capability: download.js: imports child_process and calls exec/spawn discloses shell/exec capability: index.js: imports child_process and calls exec/spawn discloses shell/exec capability: setup.js: imports child_process and calls exec/spawn discloses broad/destructive filesystem access: download.js: destructive/writing fs.* call (write/unlink/rm/chmod/rename) discloses broad/destructive filesystem access: index.js: destructive/writing fs.* call (write/unlink/rm/chmod/rename) discloses broad/destructive filesystem access: setup.js: destructive/writing fs.* call (write/unlink/rm/chmod/rename) Capability DISCLOSURE, not a verdict: static analysis sees the primitive is present and reachable, not whether its use is attacker-controlled.

• Version behavior

  fail10% of grade

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

  vs prior 3.0.379 (patch bump): new capability primitive(s) in source: child_process, fs-write, http-client; new hardcoded outbound host/BCC destination(s): download.local-mcp.com, github.com, local-mcp.com, office-mcp-production.up.railway.app, www.apple.com.

• 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 telemetry behavior — disclosure