KDPA LLC

Open source · MIT License

Open-source tools for developers who work with AI agents.

Four projects: two MCP tools on npm, a shared desktop for AI agents, and an experimental control plane for long-lived agents.

Projects

Each project lives in its own GitHub repository, with its README, security notes and issue tracker.

Local Skills MCP

Portable local skills for MCP-compatible agents and clients, with on-demand loading, validation, hot reload, multi-source discovery and trigger evaluation.

npm install -g local-skills-mcp

MCP Compression Proxy

A local MCP gateway with progressive tool discovery, context-efficient schema access through its CLI or lazy-exposure mode, shaped results, large-output controls, and aggregation of local and remote MCP servers.

npm install -g mcp-compression-proxy

agent-box Early

A shared Linux desktop with Chromium and a password vault for the AI agents on your Mac or Linux machine. Agents drive the browser with playwright-cli, and you can watch and take over from your phone.

Peers Experimental

A control plane for long-lived AI agents, with identity, durable memory, delegation, permissions, budgets and provenance.

How the projects are built

The projects are built through specification and AI agent orchestration. A maintainer sets the problem, design, constraints and acceptance bar. AI agents implement and test in scoped steps, and a maintainer reviews and accepts the result.

  • Problems, constraints, contracts, tradeoffs and limits are written down.
  • Deterministic checks are combined with adversarial or model-based evaluation where it fits.
  • Each project documents its security and execution limits. Experimental projects are labeled experimental, not presented as production-ready.

Every project uses the MIT License. Read its README and security notes before you use it with systems or data you care about.

Contact

Rafael Konrath Damaceno is one of the founders of KDPA LLC and one of the active contributors to its repositories.

For questions about a project, open an issue in its repository.