Newton’s
Menu
Create account

MODEL CONTEXT PROTOCOL (MCP) GATEWAY

Stop paying full price for MCP tool schemas.
Cut context token bills by up to 95%.

MCP servers inject extensive tool definitions, schema definitions, and resource context into every single conversation turn. Newton’s edge gateway automatically caches MCP tool schemas across turns, cutting Claude agent costs by up to 95%.

TARGET: settings.json / Extension UI

Configure Cline or Roo Code to route Anthropic & OpenAI model requests through Newton’s gateway.

CONFIGURATION SNIPPET
{
  "cline.apiProvider": "anthropic",
  "cline.anthropicBaseUrl": "https://newtons.dev",
  "cline.anthropicApiKey": "YOUR_NEWTONS_API_KEY",
  "cline.apiModelId": "claude-sonnet-5-5"
}

The MCP Tool Schema Inflation Problem

WITHOUT EDGE CACHING

Duplicate Schema Ingestion

A setup with 5 MCP servers (GitHub, Postgres, Slack, Filesystem, Browser) sends ~12,000 tool definition tokens on every single user turn. Across a 20-turn agent session, you pay for 240,000 un-cached input tokens.

WITH NEWTON’S GATEWAY

85% Prompt Cache Pass-Through

Newton’s maintains stable prefix order for tool schemas, automatically triggering Anthropic & DeepSeek prompt caching. Turns 2 through 20 get billed at up to an 85% discount.

REAL-TIME ANALYTICS

Exact Cache Hit Visibility

View exact cached token counts, read latency, and dollar savings per tool call directly in your Newton’s developer dashboard.

Frequently Asked Questions: MCP & Agent Gateways

What is an MCP gateway or agent gateway?

An MCP gateway routes tool execution requests between agentic clients (Claude Desktop, Cursor, Cline, Roo Code) and frontier model endpoints. It standardizes OpenAI and Anthropic schemas, enforces unified token budgets, and ensures prompt cache preservation across multi-turn agent conversations.

How does Newton’s reduce costs for MCP tool calls?

MCP tools inject large JSON schemas into system and tool definitions on every turn. Newton’s maintains strict prefix determinism so upstream providers (Anthropic, DeepSeek) recognize the cache block, granting prompt cache pass-through discounts of up to 85% starting from turn 2.

Can I use open-source or custom MCP servers with Newton’s?

Yes. Any MCP server running over stdio or SSE that queries an LLM upstream can use Newton’s Base URL (https://newtons.dev/api/v1) directly in its configuration without modifying SDK code.

Connect your MCP client in 30 seconds

Get your API key with $1 free starter credit — zero credit card required.

Claim $1 Starter Credit →