gcf-proxy 0.4.0

MCP proxy that re-encodes JSON tool responses as GCF. Zero code changes.

gcf-proxy

MCP proxy that re-encodes JSON tool responses as GCF — the most token-efficient wire format for LLMs. Drop-in, zero code changes. A TOON/JSON alternative that saves 63-79% of tokens.

79% fewer tokens than JSON. 34% fewer than TOON. 90.5% average comprehension accuracy across 10 models and 3 providers. 1,300+ LLM evaluations. Zero training.

Docs: gcformat.com · Proxy Guide · Playground · GCF vs TOON

Install

pip install gcf-proxy

Setup (one line change)

Before:

{"mcpServers": {"yours": {"command": "your-mcp-server"}}}

After:

{"mcpServers": {"yours": {"command": "gcf-proxy", "args": ["your-mcp-server"]}}}

Your server keeps outputting JSON. The LLM receives GCF. Nothing else changes.

Works with Claude Code, Claude Desktop, VS Code, Cursor, and any MCP client.

What it does

LLM  ←──  GCF  ←──  gcf-proxy  ←──  JSON  ←──  Your Server

1. Spawns your MCP server as a subprocess
2. Proxies stdin/stdout between client and server
3. Detects JSON payloads in tool responses
4. Re-encodes as GCF (graph profile for code intelligence, generic profile for everything else)
5. Non-convertible responses (text, HTML, errors) pass through unchanged

Savings

Tested on a real MCP tool response (10 symbols, 8 edges):

JSON   2,506 bytes  ~626 tokens
GCF      916 bytes  ~229 tokens

Savings: 63% fewer tokens

On a real agent-lsp blast_radius response (7 symbols, 47 callers):

JSON   6,515 bytes  ~1,628 tokens
GCF    4,866 bytes  ~1,216 tokens

Savings: 25% fewer tokens (generic encoding, no graph profile)

When to use

• You can't modify the server (third-party binary, another team's code)
• You want to test GCF savings without writing any code
• You want zero-effort adoption on any existing MCP server

For session deduplication (92.7% savings) and delta encoding (81.2% savings), use the GCF libraries directly.

Also available on

• npm: npm install -g @blackwell-systems/gcf-proxy
• Go: go install github.com/blackwell-systems/gcf-proxy@latest

Benchmarks

Format	Accuracy	Tokens	vs JSON
GCF	90.5% avg (10 models)	11,090	79% fewer
TOON	68.5% avg	16,378	69% fewer
JSON	53.6% avg	53,341	baseline

GCF wins all 6 datasets on TOON's own benchmark. 42% smaller on semi-uniform data, 34% on mixed-structure.

Reproduce comprehension eval: git clone https://github.com/blackwell-systems/gcf-go && cd gcf-go/eval && GOWORK=off go test -run TestComprehension -v -timeout 0

Reproduce token benchmark: git clone https://github.com/blackwell-systems/toon && cd toon && git checkout gcf-comparison && cd benchmarks && pnpm install && pnpm benchmark:tokens

Links

• Full Setup Guide
• GCF Specification
• GCF vs TOON
• Playground
• GitHub