MeetJSON Documentation

MeetJSON is a high-speed, zero-dependency Edge Optimization & Telemetry Layer for autonomous AI coding agents (Cursor, Claude Code, Windsurf, Antigravity, Aider). Operating at sub-10ms global latency across 300+ edge locations, MeetJSON deterministically compresses input context before LLM inference and auto-heals output streams without retry loops.

Quickstart (Zero-Install Remote MCP)

You do not need to install local Node.js packages or Docker daemons. MeetJSON provides an instant, universal HTTP Server-Sent Events (SSE) endpoint authenticated directly via URL query parameters.

https://meetjson.com/sse?key=YOUR_API_KEY

API Keys & Authentication

MeetJSON supports two straightforward authentication methods:

• Remote MCP (Cursor / Windsurf / Claude Desktop): Pass your key as a query parameter in the SSE URL: ?key=mj_live_...
• AI Gateway & REST API: Include the header X-MeetJSON-Key: mj_live_... or standard Authorization: Bearer mj_live_...

1. Remote MCP Protocol Integration

MeetJSON implements the official Model Context Protocol (2025-03-26 & 2026-10 specs) with built-in InitializeResult.instructions so agentic LLMs prioritize MeetJSON readers before attempting to ingest raw, uncompressed files.

Cursor IDE Configuration

Add MeetJSON to your Cursor MCP settings (~/.cursor/mcp.json or Settings → Features → MCP):

{
  "mcpServers": {
    "meetjson": {
      "url": "https://meetjson.com/sse?key=mj_live_YOUR_KEY"
    }
  }
}

Antigravity / Gemini CLI

Run a single CLI command to register MeetJSON directly into your active workspace:

agy mcp add meetjson "https://meetjson.com/sse?key=mj_live_YOUR_KEY"

Claude Code & Claude Desktop

In claude_desktop_config.json:

{
  "mcpServers": {
    "meetjson": {
      "url": "https://meetjson.com/sse?key=mj_live_YOUR_KEY"
    }
  }
}

Available MCP Tools Reference

2. High-Performance AI Gateway

Drop MeetJSON directly into any AI agent or SDK simply by changing your baseURL. MeetJSON transparently intercepts payloads, applies in-flight AST code compaction on prompt messages, auto-injects Anthropic prompt caching headers, and repairs malformed output streams.

Universal Multi-Provider Auto-Routing

MeetJSON inspects the requested model name or API key prefix and automatically routes traffic to the optimal upstream provider:

• Anthropic: /v1/messages → api.anthropic.com
• OpenAI: /v1/chat/completions → api.openai.com
• Google Gemini: /v1/gemini/* → generativelanguage.googleapis.com
• DeepSeek / Groq / xAI / OpenRouter: Automatic upstream dispatch based on model name.

from openai import OpenAI

client = OpenAI(
    base_url="https://meetjson.com/v1",
    api_key="your-openai-key",
    default_headers={"X-MeetJSON-Key": "mj_live_YOUR_KEY"}
)

response = client.chat.completions.create(
    model="gpt-6-sol",
    messages=[{"role": "user", "content": "Analyze this 2,000-line codebase: ..."}]
)

Automatic Ephemeral Prompt Caching

When routing to Anthropic (Claude Sonnet 5 / Opus 5.5), MeetJSON automatically inspects the top-level tools and system prompt blocks. If missing, it injects cache_control: { type: "ephemeral" }, instantly guaranteeing a 90% token price reduction on cached prefixes.

In-Flight Code AST Pruning

As messages transit through the gateway, large code blocks and source file contents embedded inside prompt messages are analyzed on the fly. Functions longer than 3 lines are skeletonized into signature-preserving representations, reducing prompt token size by up to 70% before hitting model context limits.

3. Core Optimization Engines

Lexical State-Machine AST Skeletonizer

Unlike naive regex scrapers that break on nested braces, template literals, or regex definitions, MeetJSON features a custom Lexical State Machine Scanner. It preserves short helper functions intact, maintains 100% of interfaces/types and comments, and folds long function bodies into exact line-range markers:

// ORIGINAL (45 lines)
export function parseTree(tokens: Token[], options?: ParserOptions): ASTNode {
  const root = new ASTNode();
  for (let i = 0; i < tokens.length; i++) {
    // 40 lines of internal loop logic...
  }
  return root;
}

// SKELETONIZED BY MEETJSON (~70% token drop)
export function parseTree(tokens: Token[], options?: ParserOptions): ASTNode {
  /* [meetjson:fold:lines 2-44 (~42 lines)] */
  return root;
}

Semantic Schema Pruner

Massive database outputs and REST API responses frequently exceed model context windows. MeetJSON’s schema pruner extracts structural shapes, deduplicates repetitive array elements, and retains representative boundary samples with accurate count placeholders—cutting JSON payload size by 85% to 95% without breaking schema validation.

// RAW INPUT (500 items, ~18,000 tokens)
[{"id": 1, "name": "Item 1", ...}, {"id": 2, "name": "Item 2", ...}, ...]

// PRUNED BY MEETJSON (~90% token reduction)
[
  {"id": 1, "name": "Item 1", "status": "active"},
  {"id": 2, "name": "Item 2", "status": "active"},
  "/* ... (+497 items omitted, schema preserved) */",
  {"id": 500, "name": "Item 500", "status": "closed"}
]

11-Stage Syntax Repair

Frontier LLMs frequently truncate tokens, emit single quotes, leave braces unclosed, or output Python literals (True, False, None) instead of valid JSON. MeetJSON’s deterministic state repairer walks through 11 edge correction passes in <1ms to ensure 100% parseable outputs before tool execution:

• Automated bracket and brace balance recovery
• Unescaped quote and control character restoration
• Python/YAML dialect normalization to strict RFC 8259 JSON
• Zero-retry tool execution guarantee

Context Time Machine

When long-running agent sessions balloon to 150,000+ tokens, agent performance degrades and latency skyrockets. MeetJSON Time Machine allows you to type magic commands into your prompt or invoke MCP checkpoints:

checkpoint save feature_auth_ready
checkpoint list
checkpoint restore feature_auth_ready

4. Edge REST API

For CI/CD pipelines (GitHub Actions, GitLab CI), backend microservices (Go, Rust, Node.js, Python), or direct shell automations, MeetJSON exposes high-speed deterministic endpoints that require zero SDK dependencies.

AST Code Skeletonizer (POST /v1/optimize-code)

Compress large repository files or PR diffs before injecting into prompt context:

curl -X POST https://meetjson.com/v1/optimize-code \
  -H "Content-Type: application/json" \
  -H "X-MeetJSON-Key: mj_live_YOUR_KEY" \
  -d '{"content": "function processOrder(id: string, items: Item[]): Order { /* 80 lines */ return order; }"}'

11-Stage Syntax Repair (POST /v1/repair)

Deterministically auto-heal truncated model outputs and broken JSON in <1ms without paying for LLM retry loops:

curl -X POST https://meetjson.com/v1/repair \
  -H "Content-Type: application/json" \
  -H "X-MeetJSON-Key: mj_live_YOUR_KEY" \
  -d '{"content": "{ status: True, items: [1, 2, "}'

Schema & Array Pruning (POST /v1/prune)

Prune massive database query results or API payloads by up to 90% while keeping schema validity intact:

curl -X POST https://meetjson.com/v1/prune \
  -H "Content-Type: application/json" \
  -H "X-MeetJSON-Key: mj_live_YOUR_KEY" \
  -d '{"content": "[{\"id\": 1, \"name\": \"User 1\"}, ... 500 items ...]"}'