Invisible characters in tool definitions or prompts

HighAV-MCP-106MCPMCP config packNew in 2.0

What it finds

Tool definitions or prompt text contain invisible Unicode characters (zero-width, bidirectional-override, or "tag" characters).

Why it matters

Invisible characters can hide instructions that the model reads but a human reviewer cannot see. This is a known way to smuggle prompt injection past code review.

What an attacker could do: Hidden text could steer the AI in ways nobody reviewing the file can see.

How to fix it

Remove all zero-width, bidi-override and tag characters from tool descriptions and prompts, and add a CI check that rejects them.

Pro shows the corrected code for your exact file, step by step, and re-scans to confirm the fix. See plans

Check details

Finding code: MCP_HIDDEN_UNICODE
How it detects: Looks for a specific risky pattern in the code or config.
Confidence: Definite: the risky setting is written plainly in the file.
Scans: MCP client configuration files: claude_desktop_config.json, .cursor/mcp.json, .vscode/mcp.json, .mcp.json
OWASP LLM Top 10: LLM01 - Prompt Injection
NIST AI RMF: MS-2.3
SOC 2: CC7.1

Check your agent for this

Paste your agent code or MCP config. The free scan runs all 124 checks in about 10 seconds — no account needed.

Scan an agent free Run it in CI or your editor