Contract validation, self-healing failover, and BYOK encryption for every MCP tool call. Drop-in middleware, zero infrastructure changes.
npx correctover-mcp-server
JSON Schema enforcement on every MCP tool input/output. Catches malformed payloads before they reach agent logic.
Hash-based response verification. Detects corrupted or tampered outputs by comparing signatures.
Semantic vector comparison against expected output profiles. Flags looks-correct-but-wrong responses.
Named entity validation against allowed schemas. Prevents PII and credential leakage through tool responses.
Regex-based pattern blocking for code execution, path traversal, and SSRF attempts in tool arguments.
Detects prompt injection attempts embedded in tool call parameters before they reach the LLM.
When a provider returns errors, Correctover transparently retries across fallback providers, validates the replacement response, and continues streaming — all within the same tool call lifecycle.
Add to your mcpServers config. Set at least one provider API key (BYOK) so the server can route tool calls:
{
"mcpServers": {
"correctover": {
"command": "npx",
"args": ["correctover-mcp-server"],
"env": {
"OPENAI_API_KEY": "sk-...",
"CORRECTOVER_LICENSE_KEY": "NB-PRO-... (optional, for Pro plan)"
}
}
}
}
Server reads provider keys from env: OPENAI_API_KEY, ANTHROPIC_API_KEY, DEEPSEEK_API_KEY, MOONSHOT_API_KEY, ZHIPU_API_KEY, DASHSCOPE_API_KEY.
Get your provider key from its dashboard (e.g. platform.openai.com/api-keys). Pro license key from rating/register.