ValidationStation

Production-grade JSON repair and validation for agent pipelines. Automatically fixes malformed LLM output using Gemini, then validates against your JSON Schema. Built for agents that need data integrity without human intervention.

$0.025 per successful call. Fully automated. No humans involved.

---

Why This Exists

LLMs generate broken JSON constantly — trailing commas, unquoted keys, truncated strings, wrong types. Passing that output raw to an API or database causes hard crashes and silent data corruption. This tool is the safety net between your generator and your consumers.

How It Works

Every call runs through three stages:

raw_input
    │
    ▼
Stage 1: Strict JSON parse      → PASS → Stage 3
    │ FAIL
    ▼
Stage 2: LLM repair (Gemini)    → re-parse → PASS → Stage 3
    │ FAIL
    ▼
REPAIR_PARSE_FAILURE error with remediation instructions
    
Stage 3: JSON Schema validation → PASS → ✅ validated_data returned
    │ FAIL
    ▼
SCHEMA_VALIDATION_FAILURE error with per-field error list

Getting Started

1. Register for an API key

curl -X POST https://mcp-json-validator-production.up.railway.app/register \
  -H "Content-Type: application/json" \
  -d '{"email": "[email protected]"}'

Returns:

{
  "api_key": "mcpjv_...",
  "checkout_url": "https://checkout.stripe.com/...",
  "billing": "Metered — $0.025 per successful call, billed monthly via Stripe."
}

Visit the checkout_url to add your payment method. Your key works immediately.

2. Connect to the MCP server

https://mcp-json-validator-production.up.railway.app/sse

3. Call the tool

{
  "tool": "sanitize_and_validate_json",
  "arguments": {
    "api_key": "mcpjv_your_key_here",
    "raw_input": "{'name': 'Alice', 'age': '30', scores: [95, 87,]}",
    "target_schema": {
      "type": "object",
      "required": ["name", "age", "scores"],
      "properties": {
        "name":   { "type": "string" },
        "age":    { "type": "integer" },
        "scores": { "type": "array", "items": { "type": "number" } }
      }
    }
  }
}

Response:

{
  "status": "valid",
  "repair_applied": true,
  "validated_data": { "name": "Alice", "age": 30, "scores": [95, 87] },
  "message": "JSON is valid and satisfies the target schema. Syntax was automatically repaired."
}

Error Codes

Code	Meaning	What to do
FORBIDDEN	Invalid or missing api_key	Register at /register
EMPTY_INPUT	raw_input was empty	Check upstream step produced output
INVALID_SCHEMA	target_schema missing or not an object	Fix schema definition
MALFORMED_SCHEMA	target_schema fails JSON Schema meta-validation	Check for typos in schema keywords
REPAIR_PARSE_FAILURE	LLM repair failed to produce valid JSON	Regenerate raw_input from scratch
REPAIR_UNAVAILABLE	Repair stage unavailable	Check server status
SCHEMA_VALIDATION_FAILURE	JSON valid but structure wrong	Read validation_errors, fix and retry

Self-Healing Loop Pattern

When you receive SCHEMA_VALIDATION_FAILURE, feed the errors back to your generator:

Your previous output had these schema violations:
<validation_errors from tool response>

Regenerate the JSON correcting every violation listed above.

Then call the tool again. Limit retries to 3 before escalating.

Endpoints

Endpoint	Method	Purpose
/sse	GET	MCP connection point
/register	POST	Get an API key
/health	GET	Server health check
/.well-known/mcp/server-card.json	GET	Smithery discovery

Pricing

$0.025 per successful call, billed monthly via Stripe. You are only charged when the tool returns status: valid. Failed calls (auth errors, schema errors) are free.

License

MIT