Skip to content

MCP_TOOL_SPECIFICATION.md

Latest commit

 

History

History
36 lines (23 loc) · 1.48 KB

MCP_TOOL_SPECIFICATION.md

File metadata and controls

36 lines (23 loc) · 1.48 KB

Encryption Module - MCP Tool Specification

Version: v1.2.3 | Status: Not Applicable | Last Updated: March 2026

Overview

The encryption module intentionally does not expose any MCP (Model Context Protocol) tools. Encryption operations involve sensitive key material and should not be invoked through an AI-facing tool interface for the following reasons:

  1. Key exposure risk: MCP tool calls may log arguments, exposing keys or plaintext to conversation transcripts.
  2. No ambient authority: Encryption keys should be managed by application code, not passed through an LLM tool call boundary.
  3. Side-effect sensitivity: File encryption/decryption and key storage have irreversible side effects that require explicit application-level authorization.

Recommended Usage

Use this module's Python API directly from application code or scripts:

from codomyrmex.encryption import AESGCMEncryptor, generate_aes_key

key = generate_aes_key()
encryptor = AESGCMEncryptor(key)
ciphertext = encryptor.encrypt(b"sensitive data")

Future Considerations

If MCP tool exposure is needed in the future, consider:

  • A hash_data tool (read-only, no key material) could be safely exposed.
  • Any tool involving keys would need a key-reference system (key IDs rather than raw bytes).
  • All tool calls should avoid logging plaintext or key material.

Navigation