ValidationStation
AI powered, self healing JSON validator. Pay/call
README
Self-Healing JSON Validator — MCP Server
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": "you@example.com"}'
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
推荐服务器
Baidu Map
百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。
Playwright MCP Server
一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。
Magic Component Platform (MCP)
一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。
Audiense Insights MCP Server
通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。
VeyraX
一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。
graphlit-mcp-server
模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。
Kagi MCP Server
一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。
e2b-mcp-server
使用 MCP 通过 e2b 运行代码。
Neon MCP Server
用于与 Neon 管理 API 和数据库交互的 MCP 服务器
Exa MCP Server
模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。