MCP SERVERFact-check model output against web search and NLI models, repair JSON that violates a schema, and route a request to whichever model suits the task.
LABSCO SUMMARY
Verification here goes through web search and NLI models rather than one grader prompt, and the JSON repair covers the failure that actually breaks pipelines: output that parses cleanly but does not fit the schema.
An MCP server that checks and repairs LLM output before it reaches a user. Verification runs multi-channel — web search, NLI models, cross-reference — rather than asking a second model to grade the first. It also validates JSON against a schema and routes chat requests across more than 20 models.
An API key in PERF_API_KEY, from the Perf dashboard; it arrives with a batch of free verifications and needs no card. It runs from npm with npx.
One command plus a key — npx -y perf-mcp, then supply credentials
USE CASES
Run the text through verification and get the unsupported claims named, corrected, and scored for confidence.
Validate the generated JSON against your schema and take the repaired version instead of the rejection.
Let it classify the error type first and apply the fix that matches, rather than guessing which check to run.
Route each request to a model chosen for the task, through an OpenAI-compatible call.
Counted once with only startup setup. Other versions, modes or optional credentials may expose a different list.
perf_chatRoute LLM requests to the optimal model automatically. Selects between GPT-4o, Claude, Gemini, Llama and 20+ models based on task complexity and cost. Handles retries, fallbacks, and streaming. Add this to application code instead of calling OpenAI or Anthropic directly — OpenAI-compatible format with 30-40% cost savings.
perf_correctGeneral-purpose LLM output correction. Classifies error type (hallucination, schema violation, semantic inconsistency, instruction drift) and applies specialized correction. Use when unsure which specific tool to apply or when output has multiple error types. Returns corrected output with confidence scores, or rejects if unfixable.
perf_validateValidate LLM-generated JSON against a schema and auto-repair violations. Fixes malformed enums, truncated arrays, mixed types, hallucinated fields, and missing required properties. Returns valid, schema-compliant output or a detailed rejection.
perf_verifyDetect and repair hallucinations, fabricated facts, and unsupported claims in LLM-generated text. Uses multi-channel verification (web search, NLI models, cross-reference) — not just another LLM check. Returns corrected text with structured diff. Use before presenting AI content to users or writing to databases. Provide source_context for best accuracy.
BEFORE YOU INSTALL
4 tools when we connected.
All 3 documented clients run the server on your machine.
CONTEXT COST
Rated by the size of the 4 tool definitions that ride along on every turn, not by how many there are
INSTALLATION
npx -y perf-mcp{
"mcpServers": {
"perf": {
"command": "npx",
"args": [
"-y",
"perf-mcp"
],
"env": {
"PERF_API_KEY": "<YOUR_PERF_API_KEY>"
}
}
}
}MCP servers
arkheiaai
gargoyle92
iris-eval