Implementations

Three official implementations, all production-ready, all MIT licensed, all zero runtime dependencies.

Go

go get github.com/blackwell-systems/gcf-go

Feature	Status
Encode	✓
Decode	✓
Session deduplication	✓
Delta encoding	✓
Thread-safe Session	✓ (sync.Mutex)
Tests	100% coverage

GitHub · pkg.go.dev · CLI

TypeScript

npm install @blackwell-systems/gcf

Feature	Status
Encode	✓
Decode	✓
Session deduplication	✓
Delta encoding	✓
ESM module	✓
Tests	34 passing

GitHub · npm · CLI

Python

pip install gcf-python

Feature	Status
Encode	✓
Decode	✓
Session deduplication	✓
Delta encoding	✓
Thread-safe Session	✓ (threading.Lock)
Type hints	✓ (full coverage)
Python 3.9+	✓
Tests	43 passing

GitHub · PyPI · CLI

MCP Proxy

Drop-in wrapper for any existing MCP server. Zero code changes required.

go install github.com/blackwell-systems/gcf-proxy@latest

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

JSON responses are re-encoded as GCF mid-flight. Your server keeps outputting JSON; the LLM receives GCF.

GitHub · pkg.go.dev

Output compatibility

All three implementations produce byte-for-byte identical output for the same input. The format is the product, not the implementation. Token efficiency numbers are independent of which library you use.

Contributing an implementation

GCF is simple enough to implement in a weekend. The spec is 229 lines of EBNF + prose. If you're building an implementation:

1. Start with Encode (most useful, simplest)
2. Add Decode (needed for testing)
3. Add Session (needed for production MCP servers)
4. Add Delta (needed for high-frequency re-queries)

Validate against the conformance test suite (14 fixtures across encode, decode, session, delta, and error cases). Test against the comprehension eval to verify your output is LLM-comprehensible at scale.