Skip to main content
Glama

Koma Gate MCP

Koma Gate のプロンプトインジェクション分類を、AI エージェント向けの MCP ツールとして公開します。

npm install koma-gate-mcp

機能

AI エージェントが信頼できないユーザー入力に対して、動作前に安全性とスコープ内であるかを確認できる単一のツール classify_input を提供します。

Related MCP server: jamjet-policy

セットアップ

プロバイダーと API キーを設定します。

# Provider: openai | anthropic | google | deepseek | ollama (default: google)
export KOMA_PROVIDER=google
export GEMINI_API_KEY=sk-...

Claude Desktop

claude_desktop_config.json に以下を追加します。

{
  "mcpServers": {
    "koma-gate": {
      "command": "npx",
      "args": ["-y", "koma-gate-mcp"]
    }
  }
}

ツール: classify_input

  • text (必須) — 分類する信頼できないユーザー入力

  • preset (オプション) — general | code | support | reference (デフォルト: general)

戻り値:

{
  "allowed": true,
  "in_scope": true,
  "reason": "in scope",
  "preset": "general",
  "model": "gemini-2.5-flash"
}

セキュリティ境界

これは LLM ベースのスコープ分類器であり、暗号学的なプロンプトインジェクション防御ではありません。完全なセキュリティ境界については koma-gate の README を、評価結果については BENCHMARKS.md を参照してください。

Available Tools

1 tool
classify_inputA

Check whether user input is safe and in-scope before an AI system processes it. Detects prompt injection, jailbreak attempts, instruction overrides, and off-topic requests. Use this to validate untrusted user text before passing it to an LLM or executing a tool.

Inputs: text (required), preset.

Tool Schema Changelog

Recent tool additions, removals, and schema changes observed during successful MCP inspections.

Loading tool schema changelog…

    TDQS

    A4.1/5.0

    Scored across 1 tool

    Disambiguation5/5

    With only one tool, there are no possible ambiguities. The single tool 'classify_input' has a clear, distinct purpose.

    Naming Consistency5/5

    The single tool name 'classify_input' follows a consistent verb_noun pattern (verb 'classify' + noun 'input'), which is clear and predictable.

    Tool Count4/5

    One tool is minimal but appropriate for a narrowly scoped server focused on a single classification task. It's not trivial—it serves a specific, valuable function as a gatekeeper.

    Completeness4/5

    The tool covers the core purpose of input safety classification. Minor gaps might include configuration options or status reporting, but the essential functionality is present.

    Maintenance

    ActivitySlowing
    ResponsivenessNo recent community issues

    Related MCP Connectors

    Related MCP Servers

    • A
      license
      Not graded
      quality
      D
      maintenance
      Adaptive security for AI agents: assess inputs for prompt injection, scan outputs for credential/PII leaks, teach new attack patterns to semantic memory, harden prompts, and monitor metrics. Runs locally via MCP stdio.
      120 PyPI
      Apache 2.0
    • A
      license
      A
      quality
      C
      maintenance
      Drop-in stdio interceptor that gates MCP tools/call requests through a YAML policy (block / require_approval / audit / budget cap) before they reach the real server. The same policy file is reused by @jamjet/claude-code-hook and @jamjet/openai-guardrail, so one rule set covers Claude Desktop, Cursor, OpenAI Agents, and custom clients.
      3
      2
      Apache 2.0
    • F
      license
      Not graded
      quality
      B
      maintenance
      Enables autonomous agents and MCP-compatible clients to route inputs through a deterministic firewall that flags and sanitizes prompt injection, jailbreaks, system prompt exfiltration, and tool hijacking. It supports JSON-RPC stdio integration for tools like Claude Desktop, Cursor, and multi-agent frameworks.
      7
      -