Docs

MCP server for AI agents

An agent cannot drive a browser interface. The MCP server exposes the same engine as callable tools, so an assistant can format, validate and type a document as a step in its own work.

Endpoint

https://jsonparse.online/mcp

Streamable HTTP transport, stateless, no authentication. Every request is self-contained, so there is no session to keep alive and no state to lose between calls.

Connecting

Claude Code

claude mcp add --transport http jsonparse https://jsonparse.online/mcp

Any client with a JSON config

{
  "mcpServers": {
    "jsonparse": {
      "type": "http",
      "url": "https://jsonparse.online/mcp"
    }
  }
}

Tools

Tool	Input	Returns
`format_json`	`json`, `mode`, `indent`	Beautified or minified text.
`validate_json`	`json`	Valid, or the first error with line, column and offset.
`json_to_typescript`	`json`, plus the generator options	TypeScript interfaces or type aliases.
`json_to_schema`	`json`, plus the schema options	A JSON Schema document.

What it is good for

Turning a captured API response into types without the agent hand-writing them and getting the optional fields wrong.
Locating the syntax error in a file an agent generated, precisely, instead of guessing from a parser message.
Producing a schema from a sample as a step in a larger task.

The same boundary as the HTTP API applies: documents sent to the MCP server reach the server. Documents opened in the viewer never do.

Frequently asked questions

Does the MCP server need an API key?

No. It is open and rate limited by address, the same as the HTTP API.

Which MCP clients work with it?

Any client supporting the Streamable HTTP transport, which includes Claude Code, the Claude desktop app via a remote server entry, and the common agent frameworks.

Can I run it myself?

Yes. It is part of the same MIT-licensed repository and deploys as a single serverless function.

Last updated 2026-09-14.