prompt-protection MCP server

Read from the package source without running it. The installed server may list more.

• check_tool_call

  Check a model-proposed tool call against provenance policy before executing it: blocks when untrusted data (URLs, emails, paths, verbatim payloads) flows into a network / email / exec / file / payment sink, flags when an injection-scored source preceded it. Pass `sources` for a stateless check; otherwise uses sources registered with register_source. `userText` marks destinations the user named as

• detect_canary

  Check model output for a leaked system-prompt canary token in any form (exact, spaced, base64, hex, reversed, partial). Optionally compares the output against the system prompt for verbatim overlap.

• register_source

  Register an untrusted tool result (email, web page, file, RAG chunk) with the tool-call guard so later check_tool_call decisions can trace data flowing out of it. Returns the injection scan of the source.

• scan_output

  Scan model output for signs of compromise: leaked system prompts, exposed credentials, PII, or relayed injection aimed at a downstream consumer.

• scan_prompt

  Scan untrusted user text for prompt injection, jailbreaks, and data exfiltration before it reaches the model. Returns a verdict (allow / flag / block).

• scan_tool_definition

  Scan a tool/function definition (name, description, parameter schema) for tool poisoning, hidden instructions, concealment directives, or exfiltration embedded in the metadata. Call before trusting a newly discovered tool.

• Code scan66 source files scanned25/25
• –Live reliabilityno gateway calls yet and no remote to proben/a
• –Tool poisoningtools not inspected (local package is not executed); not countedn/a
• Auth qualitylocal package, no credentials required12/15
• Maintenancelast push 1 days ago15/15
• Maintainer identityregistry namespace matches repository owner; GitHub account older than a year8/10

prompt-protection: common questions

Is prompt-protection MCP server safe?

Yes, by our scan: it is graded A (92/100). Read the prompt-protection safety report

How do I install prompt-protection?

It runs on your machine. Copy the Claude Code, Claude Desktop or Cursor config from the install section.

Does prompt-protection need an API key?

Not as far as the registry entry and our scan can tell: no credentials are declared or required.

Is prompt-protection maintained?

The last commit was in the last day (2026-09-18). The latest release is v3.1.1.

What can I use instead of prompt-protection?

Servers from other publishers that do the same job: mcpm MCP server, Aletheia Runtime Safety & Scope Creep Filter MCP server and Agent-Native Analytics MCP server. Compare all prompt-protection alternatives.