ExMCP.Content.Sanitizer (ex_mcp v1.0.0-rc.8)

Content sanitization utilities for MCP content.

Experimental

Not part of the core stable Handler/DSL surface. Best-effort text sanitization only — not a full security sandbox.

Implemented

:html_escape / html_escape/1
:strip_scripts / strip_scripts/1
:normalize_unicode / normalize_unicode/1 (Unicode NFC)
:limit_size (text truncation; binary content marked on overflow)
sanitize_path/1, strip_sql_injection/1

Deprecated (not MCP/ACP requirements)

:remove_metadata / remove_metadata/1 — EXIF stripping was never implemented; not required by MCP. Prefer app-level media pipelines if needed.
:compress_media — no-op stub; planned for removal in 2.0.0

Summary

Types

sanitization_op()

Sanitization operation

Functions

html_escape(text)

Escapes HTML entities in text content.

normalize_unicode(text)

Normalizes Unicode text to NFC form.

remove_metadata(content) deprecated

Removes potentially dangerous metadata from content.

sanitize(content, operations)

Sanitizes content by applying a list of sanitization operations.

sanitize_path(path)

Sanitizes file paths to prevent directory traversal.

sanitize_text(text, operations)

Sanitizes text content specifically.

strip_scripts(text)

Removes script tags and JavaScript from HTML content.

strip_sql_injection(text)

Removes SQL injection attempts from text.

Types

sanitization_op()

@type sanitization_op() ::
  :html_escape
  | :strip_scripts
  | :normalize_unicode
  | :limit_size
  | :remove_metadata
  | :compress_media
  | {:custom, function()}
  | atom()

Sanitization operation

Functions

html_escape(text)

@spec html_escape(String.t()) :: String.t()

Escapes HTML entities in text content.

normalize_unicode(text)

@spec normalize_unicode(String.t()) :: String.t()

Normalizes Unicode text to NFC form.

Uses :unicode.characters_to_nfc_binary/1. This reduces some homograph confusion but is not a complete security control by itself.

remove_metadata(content)

This function is deprecated. Not implemented for EXIF; not an MCP requirement. Planned for removal in 2.0.0..

@spec remove_metadata(ExMCP.Content.Protocol.content()) ::
  ExMCP.Content.Protocol.content()

Removes potentially dangerous metadata from content.

Deprecated

Never implemented for real EXIF stripping. Not required by MCP/ACP. Returns content unchanged (or clears a :metadata key when present). Planned for removal in 2.0.0.

sanitize(content, operations)

@spec sanitize(ExMCP.Content.Protocol.content(), [sanitization_op()]) ::
  ExMCP.Content.Protocol.content()

Sanitizes content by applying a list of sanitization operations.

Examples

safe_content = Sanitizer.sanitize(content, [
  :html_escape,
  :strip_scripts,
  {:limit_size, 1_000_000}
])

sanitize_path(path)

@spec sanitize_path(String.t()) :: String.t()

Sanitizes file paths to prevent directory traversal.

sanitize_text(text, operations)

@spec sanitize_text(String.t(), [sanitization_op()]) :: String.t()

Sanitizes text content specifically.

strip_scripts(text)

@spec strip_scripts(String.t()) :: String.t()

Removes script tags and JavaScript from HTML content.

strip_sql_injection(text)

@spec strip_sql_injection(String.t()) :: String.t()

Removes SQL injection attempts from text.