MCP 服务器

perf-mcp

Fact-checks and fixes AI outputs by catching hallucinations, repairing broken JSON, and correcting errors before they reach users, with tools for verification, validation, and correction.

README

perf-mcp

Fact-check and fix AI outputs. Catches hallucinations, repairs broken JSON, corrects errors — before they reach users.

Works with Claude Code, Cursor, Windsurf, Cline, and any MCP-compatible client.

Quick Start

Add to your MCP client config:

{
  "mcpServers": {
    "perf": {
      "command": "npx",
      "args": ["-y", "perf-mcp"],
      "env": {
        "PERF_API_KEY": "pk_live_xxx"
      }
    }
  }
}

Get your API key at dashboard.withperf.pro — 200 free verifications, no credit card.

Tools

`perf_verify`

Detect and repair hallucinations in LLM-generated text. Uses multi-channel verification (web search, NLI models, cross-reference) — not just another LLM check.

perf_verify({ content: "The Eiffel Tower was built in 1887." })
→ Corrected: "built in 1887" → "inaugurated in 1889" (89% confidence)

`perf_validate`

Validate LLM-generated JSON against a schema and auto-repair violations. Fixes malformed enums, wrong types, missing fields, hallucinated properties.

perf_validate({
  content: '{"name": "John", "age": "twenty"}',
  target_schema: { type: "object", properties: { name: { type: "string" }, age: { type: "number" } } }
})
→ Rejected: /age must be number

`perf_correct`

General-purpose output correction. Classifies the error type and applies the right fix — hallucination, schema violation, semantic inconsistency, or instruction drift.

perf_correct({ content: "The Great Wall was built in 1950.", correction_budget: "fast" })
→ Corrected: temporal_error + factual_error detected (87% confidence)

`perf_chat`

Route LLM requests to the optimal model automatically. Selects between GPT-4o, Claude, Gemini, and 20+ models based on task complexity. OpenAI-compatible format.

Setup by Client

Claude Code

Add to your project's .mcp.json:

{
  "mcpServers": {
    "perf": {
      "command": "npx",
      "args": ["-y", "perf-mcp"],
      "env": {
        "PERF_API_KEY": "pk_live_xxx"
      }
    }
  }
}

Cursor

Settings → MCP → Add Server:

{
  "mcpServers": {
    "perf": {
      "command": "npx",
      "args": ["-y", "perf-mcp"],
      "env": {
        "PERF_API_KEY": "pk_live_xxx"
      }
    }
  }
}

Windsurf

Add to ~/.codeium/windsurf/mcp_config.json:

{
  "mcpServers": {
    "perf": {
      "command": "npx",
      "args": ["-y", "perf-mcp"],
      "env": {
        "PERF_API_KEY": "pk_live_xxx"
      }
    }
  }
}