@Rih0z
基本情報
Adversarial AI review API — independent AI reviews another AI's output. Stop LLMs from grading their own homework. Provides automated quality assurance for AI-generated code, content, and other outputs through independent review pipelines.
以下の設定を使って、このサーバーを MCP 対応クライアントに追加してください。
{
"mcpServers": {
"agentdesk-mcp": {
"command": "npx",
"args": [
"-y",
"@ezark-publish/agentdesk-mcp@1.3.0"
]
}
}
}Adversarial quality review of any AI-generated output. An independent reviewer assumes the author made mistakes and actively looks for problems. Returns structured verdict (PASS/FAIL/CONDITIONAL_PASS), score (0-100), categorized issues with severity, and evidence-based checklist. Works for any output type: code, content, summaries, translations, data extraction, etc.
Dual adversarial review: two independent reviewers assess the output from different angles, then a merge agent combines their findings. Stricter than single review — if either reviewer finds a critical issue, the merged verdict is FAIL. Use for high-stakes outputs where quality is critical.
List all available services on the AgentDesk marketplace. Returns service catalog with pricing, quality scores, and capabilities. Filter by category, minimum quality score, maximum price, or capability.
Execute a service on the AgentDesk marketplace. Requires an AgentDesk API key for authentication. Pass service-specific input parameters.
AgentDesk MCP is an MCP tool that provides independent adversarial quality review for any AI-generated output. It integrates with Claude Code, Claude Desktop, and any MCP client, and uses your own Anthropic API key (BYOK) to perform structured, bias-resistant evaluations.
Install via npx agentdesk-mcp and add it to your MCP client, setting the ANTHROPIC_API_KEY environment variable. Two tools are provided: review_output for a single adversarial review, and review_dual for a dual‑reviewer pass with a merged verdict. Pass the AI output to review, optionally customizing criteria and review type.
Self‑review has a systematic leniency bias because the same model shares blind spots that created errors. AgentDesk uses a separate reviewer invocation with adversarial prompting, which is fundamentally different and more rigorous.
Dual review (`review_dual`) runs two independent adversarial reviews from different angles, then a merge agent combines findings. If either reviewer finds a critical issue, the merged verdict is FAIL, and the lower score is taken. Issues are combined and deduplicated.
Like all LLM‑as‑judge systems, it can be vulnerable to prompt injection. The anti‑gaming validation layer mitigates superficial gaming, but determined adversarial inputs remain a challenge. Each `review_output` call makes 1 LLM API call; `review_dual` makes 3, so factor BYOK cost into your pipeline.
You need an `ANTHROPIC_API_KEY` environment variable. The default reviewer model is Claude Sonnet 4‑6, but you can specify another model. No other SDK or dependencies are required.
AgentDesk MCP offers one‑tool MCP setup, adversarial review, dual reviewer capability, and anti‑gaming validation — features not found in manual prompts, Braintrust, or DeepEval. It is MCP native and requires no SDK.
Console, Backend and CLI to manage Capgo Instant update and Native build for Capacitor apps Build and upload to app store your mobile app
Agent-Native Terminal
Know when your AI agents, cron jobs and CI pipelines fail, stall or go quiet. Your agent sets up monitors from a plain description, tells you what is broken and why, reads run history and traces, and writes notes on inci
MCP server for AI Agents — exposes Yocoolab feedback threads and design selections as tools
コメント