Skip to content

security-model.md

Latest commit

 

History

History
30 lines (19 loc) · 1.02 KB

security-model.md

File metadata and controls

30 lines (19 loc) · 1.02 KB

Security Model

Türkçe sürüm: docs/tr/security/security-model.md

mcp-code uses a deny-by-default posture around mutation and verification.

Security boundaries

  • The public tool surface does not expose raw file CRUD helpers.
  • The public tool surface does not expose arbitrary shell execution.
  • Edit application only runs against a remembered edit plan.
  • Verification only uses configured script names and allowed commands.
  • Workspace access stays bound to configured allowed paths.

Mutation guardrails

apply_safe_edit requires a valid planId produced by plan_safe_edit.

That protects the workflow in two ways:

  1. The edit must fit within previously planned symbol and file scope.
  2. The analyzer can reject edits that do not match the approved plan boundary.

Logging and protocol separation

  • MCP protocol traffic uses stdout.
  • Logs are written to stderr.
  • In-memory telemetry is used for runtime sampling without changing the public contract.

Last updated: 2026-03-10