---
title: "Cut MCP Token Usage & Context Bloat — onemcp"
description: "MCP servers resend hundreds of tool definitions every request. onemcp's Code Mode replaces them with three meta-tools — about 90% less overhead per call."
url: "https://onemcp.dev/use-cases/reduce-token-usage/"
---

# Stop MCP tool definitions from eating your context.

Every MCP server floods the prompt with its full tool schema on every request. onemcp exposes just three meta-tools and lets the model script against everything else — so the context window goes to your problem, not to boilerplate.

## How onemcp shrinks the overhead

1. **Aggregate behind a portal** — onemcp connects to all your servers and namespaces their tools on one endpoint.
2. **Expose three meta-tools** — search, describe, and execute replace every up-front tool definition.
3. **Let the model write code** — Cloudflare Code Mode runs one script that calls only the tools it actually needs.

## What you get back

- **~90% smaller system prompt** — Tool overhead drops from roughly 15,000 tokens to about 500 per request.
- **Lower cost and latency** — Fewer tokens in means cheaper, faster turns — on every single call.
- **Tools on demand** — Signatures are fetched only when needed, not stuffed in up front.

## Give the context window back to your agent.

Put your servers behind one portal and let Code Mode do the rest.