ProDisco

What it does

ProDisco implements Anthropic's Progressive Disclosure pattern for MCP servers. Instead of exposing every possible tool at once, it gives AI agents two focused tools: one to search API documentation extracted from TypeScript declaration files, and one to execute code in an isolated sandbox. The result is low token usage, fast responses, and no large payloads passing through the model.

While the primary example configuration targets Kubernetes (via @kubernetes/client-node), Prometheus, and Loki, ProDisco is a generic framework — you can point it at any TypeScript library and get instant API discoverability and sandboxed execution.

Key features

• AST-powered API indexing — extracts method signatures, types, and return types directly from .d.ts files; stays current as dependencies are upgraded
• Isolated sandbox execution — each session gets its own environment; supports local subprocess or Kubernetes CRD (Kata VM) for strong isolation
• Progressive disclosure — prodisco.searchTools surfaces only what the agent asks for; prodisco.runSandbox runs code and returns only console output
• Multi-transport — stdio (default) or HTTP with SSE streaming; supports TLS/mTLS for production deployments
• Advanced analytics — ships with simple-statistics, ml-regression, mathjs, and fft-js for anomaly detection, forecasting, and correlation analysis

Installation

Claude Desktop / Claude Code (Kubernetes):

curl -O https://raw.githubusercontent.com/harche/ProDisco/main/examples/prodisco.kubernetes.yaml
claude mcp add ProDisco --env KUBECONFIG="${HOME}/.kube/config" -- npx -y @prodisco/mcp-server --config prodisco.kubernetes.yaml

Claude Desktop config (claude_desktop_config.json):

{
  "mcpServers": {
    "ProDisco": {
      "command": "npx",
      "args": ["-y", "@prodisco/mcp-server", "--config", "prodisco.kubernetes.yaml"],
      "env": { "KUBECONFIG": "/path/to/.kube/config" }
    }
  }
}

Supported hosts

Confirmed in README: Claude Code and Claude Desktop.