Skip to main content
plugin CX Codex

gcf-proxy — Claude Code plugin: what it does, how to install

gcf-proxy wraps MCP servers to re-encode JSON tool responses into GCF, cutting context token usage.

Apache-2.0 ★ 1,596 updated 2026-10-29

Install into Claude Code · Plugin marketplace All methods and platforms ↓
/plugin marketplace add hashgraph-online/awesome-codex-plugins
/plugin install gcf-proxy@awesome-codex-plugins
Open on GitHub

What gcf-proxy does

Standard MCP tool calls return raw JSON. This plugin places a proxy between the MCP server and the model to re-encode responses into GCF format. Benchmarks report a 71% token reduction on initial calls and up to 92% on warm sessions via structural deduplication.

Setup runs through /gcf-proxy:setup <server>. The command adjusts your MCP configuration to route traffic via the proxy, preserving the disabled original entry with a -raw suffix. Included hooks clear metrics on session start and report saved token counts upon stopping. The underlying server needs no code modifications because it keeps emitting standard JSON.

What it helps with

  • reduce context token consumption from verbose MCP tool responses
  • route an existing MCP server through a compression proxy without editing server code
  • track session token savings and rewritten calls in real time

Example prompts for the agent

  • /gcf-proxy:setup github
  • /gcf-proxy:stats
  • Route the postgres server through the proxy using /gcf-proxy:setup postgres

How to install gcf-proxy

Plugin marketplace

/plugin marketplace add hashgraph-online/awesome-codex-plugins
/plugin install gcf-proxy@awesome-codex-plugins

Step by step, with commands for each platform: Claude Code plugins: marketplaces, installing, and building your own

Questions about gcf-proxy

How do I install gcf-proxy in Claude Code?

Run both commands in order: /plugin marketplace add hashgraph-online/awesome-codex-plugins followed by /plugin install gcf-proxy@awesome-codex-plugins.

What does gcf-proxy do?

It intercepts JSON responses from MCP servers and re-encodes them into GCF before passing them to the language model.

Does wrapping a server break the existing MCP configuration?

No, the setup command preserves your original server entry under a -raw suffix in disabled mode, making it possible to revert at any point.

Source text

gcf-codex-plugin · Apache-2.0
# gcf-codex-plugin

Codex plugin for [GCF Proxy](https://github.com/blackwell-systems/gcf-proxy). Reduces MCP tool call token costs by 71%.

## Install

```bash
codex plugin add gcf-proxy
```

## What it does

Every MCP tool call returns JSON. This plugin wraps your MCP servers with [GCF Proxy](https://github.com/blackwell-systems/gcf-proxy), which re-encodes those JSON responses as GCF before they reach the model.

**The savings:**

| Metric | Value |
|--------|-------|
| Token reduction vs JSON | **71%** |
| Token reduction vs TOON | **25%** |
| Session dedup (warm session) | **up to 92%** |
| Comprehension accuracy | **100%** across 1,700+ LLM evaluations |

## Skills

| Skill | Description |
|-------|-------------|
| `/gcf-proxy:setup <server>` | Wrap any MCP server with gcf-proxy |
| `/gcf-proxy:stats` | Show token savings for the current session |

## Hooks

- **SessionStart**: Clears stale stats from previous sessions
- **Stop**: Shows notification with calls rewritten, % saved, and tokens saved

## Quick start

After installing, wrap any MCP server:

```
/gcf-proxy:setup github
```

This modifies your MCP config to route the server through gcf-proxy. The original config is preserved (disabled, with a `-raw` suffix) so you can revert anytime.

## How the savings work

1. **First call (71% savings):** GCF encodes the same structured data in 71% fewer tokens than JSON. The model reads GCF natively with 100% comprehension accuracy.
2. **Subsequent calls (up to 92%):** Session deduplication detects repeated structure. Only deltas are transmitted.
3. **Zero code changes:** The proxy wraps any MCP server. The server still outputs JSON; the proxy re-encodes before it reaches the model.

## Cost example

A team running 1,000 queries/day with GPT-5.5 ($5/MTok input):

| Format | Monthly cost | Annual cost |
|--------|-------------|-------------|
| JSON | $12,098 | $145,171 |
| GCF | $3,549 | $42,583 |
| **Savings** | **$8,549/mo** | **$102,588/yr** |

[Try the cost calculator](https://gcformat.com/calculator) with your own numbers.

## Links

- [GCF Spec](https://gcformat.com)
- [GCF Proxy](https://github.com/blackwell-systems/gcf-proxy)
- [Claude Code Plugin](https://github.com/blackwell-systems/gcf-claude-plugin)
- [Benchmarks](https://gcformat.com/guide/benchmarks.html)
- [Cost Calculator](https://gcformat.com/calculator)

## License

MIT

© Blackwell Systems · Apache-2.0 · hashgraph-online/awesome-codex-plugins

Similar plugins

All plugins for Agent setup: memory, context, orchestration →

From the same set391

The whole set →
hera-godot hera-godot connects your agent to a running Godot 4.x editor through the Hera CLI, so scene edits, signals, play mode and runtime checks all happen from chat. CX Codex ★ 1,596repo stars graymatter graymatter supplies persistent memory and shared graph context for Claude Code and Codex agents over MCP. CX Codex ★ 1,596repo stars codex-goal-progress codex-goal-progress is a plugin for Claude Code and Codex that shows progress for a Codex Goal and token usage. Enable it separately for each new Goal. CX Codex ★ 1,596repo stars generative-media-skills generative-media-skills — a plugin that gives your agent 13 skills for image, video and audio generation across 100+ models, from FLUX and Midjourney v7 to Suno. CX Codex ★ 1,596repo stars frappe-agent Frappe and ERPNext guidance for a coding assistant: bench checks, DocType design, SQL choices, customization layers and ecosystem apps. CX Codex ★ 1,596repo stars figma-maxxing The figma-maxxing plugin for Claude Code and Codex provides eight skills for editing live Figma files, with pre-write checks, property read-back and comment fixes. CX Codex ★ 1,596repo stars falsegreen-skill falsegreen-skill detects false-positive test suites in Python, TypeScript, JavaScript, and Robot Framework using six ordered LLM semantic judgments. CX Codex ★ 1,596repo stars espresso Espresso steers the model toward concise replies and smaller code changes, with Ponytail bundled. It suits project research and focused edits. CX Codex ★ 1,596repo stars

Project sponsors

Companies that keep this analytics open