GCF: The AI-Native Wire Format

50-92% fewer tokens than JSON. 100% comprehension on every frontier model. Zero training required.

GCF (Graph Compact Format) replaces JSON as the data encoding between your tools and your LLM. Same data, dramatically fewer tokens, better model accuracy. This extension adds syntax highlighting for .gcf files in VS Code.

Why GCF?

Every time an MCP server, agent, or tool sends structured data to an LLM, JSON wastes 52% of tokens on repeated field names and 29% on structural characters. At 500 rows, 81% of JSON tokens carry zero information.

GCF declares field names once in a header. Rows are positional pipe-separated values. Nested objects flatten into path columns ("customer>name"). The result:

GCF profile=generic
## orders [3]{id,"customer>name","customer>email",total,status}
ORD-1|Alice|alice@co.com|99.99|shipped
ORD-2|Bob|bob@co.com|49.99|pending
ORD-3|Carol|carol@co.com|29.99|processing

vs the equivalent JSON (3x more tokens):

[{"id":"ORD-1","customer":{"name":"Alice","email":"alice@co.com"},"total":99.99,"status":"shipped"},
 {"id":"ORD-2","customer":{"name":"Bob","email":"bob@co.com"},"total":49.99,"status":"pending"},
 {"id":"ORD-3","customer":{"name":"Carol","email":"carol@co.com"},"total":29.99,"status":"processing"}]

The numbers

50-92% fewer tokens than JSON (varies by data complexity and session reuse)
100% comprehension on every frontier model (Claude, GPT-5.5, Gemini, Grok)
15/16 wins vs TOON across 16 real-world datasets
43 billion+ lossless round-trips verified across 17 serialization formats
2,400+ LLM evaluations across 11 models, 3 providers
Zero training. Models read GCF natively. No fine-tuning, no few-shot examples.

Syntax highlighting

This extension highlights all GCF constructs:

Headers: GCF profile=generic, GCF profile=graph tool=...
Section headers: ## name [N]{field1,field2}
Flattened paths: "customer>name", "billing>address>city"
Tabular rows: pipe-separated values with positional columns
Attachments: ^, ^{fields}, .fieldname {}
Graph symbols: @0 fn pkg.Auth 0.78 lsp
Edges: @0<@1 calls
Scalars: numbers, booleans, null (-), absent (~), quoted strings
Comments: # comment text

Get started

pip install gcf-python                    # Python
npm install @blackwell-systems/gcf        # TypeScript
go get github.com/blackwell-systems/gcf-go  # Go
cargo add gcf                             # Rust

Or wrap any existing MCP server with zero code changes:

pip install gcf-proxy                     # PyPI
npm install -g @blackwell-systems/gcf-proxy  # npm
go install github.com/blackwell-systems/gcf-proxy@latest  # Go

Documentation: gcformat.com Playground: gcformat.com/playground Specification: Spec v3.2 Stable Benchmarks: gcformat.com/guide/benchmarks

Adopted by

Speakeasy (customers: Google, Verizon, Mistral AI, DocuSign, Vercel) · OmniRoute (6.1K stars) · NeuroNest · Open Data Products SDK (Linux Foundation) · and more

GCF - Graph Compact Format

Blackwell Systems