Neura Market
GCF Proxy logo

GCF Proxy

Free

MCP 工具响应压缩代理。零代码接入，将任意 MCP 服务器的 JSON-RPC 响应自动转换为紧凑的 GCF 格式以节省 token，支持 JSON、YAML、TOML、CSV、MessagePack 多种格式。

FreeFree tier
Type
Open Source
Visit Website GitHub (40 stars)

About GCF Proxy

GCF Proxy is a bidirectional MCP (Model Context Protocol) proxy that translates between JSON and GCF (General Compact Format) to reduce token usage and cut costs when calling tools from LLM clients. It wraps any MCP server with zero code changes on either side—simply add 'gcf-proxy' to your MCP command. The proxy re-encodes structured JSON responses into the compact GCF format mid-flight, achieving 53–71% fewer tokens compared to raw JSON, with 100% comprehension on all leading frontier models (2,400+ evals across 11 models and 3 providers). It supports nested object flattening with an opt-out for open-weight models, session deduplication, response caching, delta encoding, and configurable minimum size and streaming thresholds. GCF Proxy can be used in stdio mode (local servers), HTTP mode (remote Streamable HTTP servers), or deployed as a standalone HTTP service, and it works with any structured data format. Installation is available via pip, npm, or Go.

Key Features

Drop-in proxy – no changes to MCP server or client
Converts JSON-RPC responses to compact GCF format automatically
53–71% fewer tokens than JSON (2,400+ evals, 11 models, 3 providers)
Nested object flattening with opt-out for open-weight models
Session deduplication (bare refs for repeated symbols)
Response caching for identical tool calls
Delta encoding – sends only changed symbols
Configurable minimum response size for encoding
Streaming mode with configurable symbol threshold
Works with stdio, HTTP, and Streamable HTTP transports

Pros & Cons

Pros
  • Significant token savings (53–71%) without quality loss
  • Zero configuration – one line change in MCP config
  • Supports multiple transports (stdio, HTTP, Streamable HTTP)
  • Bidirectional – handles both response encoding and argument decoding
  • Works with any structured data and any MCP server
  • Proven comprehension on all major frontier models
  • Open source and free to use
Cons
  • Adds a proxy layer which may introduce slight latency
  • Primarily designed for MCP ecosystem; not useful outside MCP contexts
  • Open-weight models may require --no-flatten for best comprehension (still outperforms JSON)
  • Not beneficial for very small responses (min-size flag default 100 bytes)
  • Limited documentation beyond GitHub README

Best For

Reduce token costs when using MCP-based tools with LLMsOptimize structured data responses from financial data APIs (e.g., Yahoo Finance)Deploy as a token-saving proxy for any MCP server in productionEnable cost-efficient tool calling in frontier model applicationsWork with any MCP client needing compression of JSON-RPC responses

FAQ

What is GCF Proxy?
GCF Proxy is a bidirectional MCP proxy that translates JSON-RPC responses from MCP servers into a compact format called GCF (General Compact Format) to reduce token usage, with no changes needed to your server or client.
How much can I save on tokens?
Independent evals show 53–71% fewer tokens compared to raw JSON, tested on 2,400+ evals across 11 models and 3 providers. A live example with Yahoo Finance data saved 46.2% bytes and ~13,600 tokens in three tool calls.
Is it compatible with any MCP server?
Yes, GCF Proxy works with any standard MCP server. Simply add 'gcf-proxy' in front of the MCP command for stdio mode, or use --upstream for HTTP servers. It supports JSON and SSE responses, and session ID tracking via Mcp-Session-Id is automatic.
Do I need to change my client or server code?
No changes are needed. The proxy is a drop-in wrapper that translates mid-flight. Both sides continue to use standard JSON-RPC; the proxy handles compression and decompression transparently.
How do I install GCF Proxy?
Install via pip: 'pip install gcf-proxy', via npm: 'npm install -g @blackwell-systems/gcf-proxy', or via Go: 'go install github.com/blackwell-systems/gcf-proxy@latest'.