I read Anthropic and Cloudflare latest articles about AI agents struggling with context bloat and agents bad execution using MPCs and how they excel at executing tasks using Typescrip. I built CodeModeTOON for my own workflows and it's been solid so far so i decided to publish it.
What it does: - TOON compression: Extracts schema from structured JSON, compresses values. Gets 30-90% savings on structured data (K8s manifests, logs, API responses). - Lazy loading: MCP servers only start when needed. - Pre-built workflows: Research, K8s auditing, incident analysis.
limitations: - Unstructured text compresses poorly (~4%). - Uses Node.js vm module, so not suitable for multi-tenant deployments.
It's MIT licensed.
GitHub: https://github.com/ziad-hsn/code-mode-toon
- https://www.anthropic.com/engineering/code-execution-with-mc... - https://blog.cloudflare.com/code-mode/
Happy to answer any questions about the implementation and feedback will be very appreciated.