describe
KeppyLab
AI-native capability management for MCP tool stacks.
describe helps an AI client discover MCP servers, install or register the right ones, write client configuration, and read the resulting capability map back as resources and prompts.
Why It Exists
AI clients are no longer just chat windows with a few tool buttons. They need a living map of capabilities: which tools exist, which are installed, which are safe to call, which need credentials, and which context resources can help the model choose the next action.
describe gives that map to the model.
What Changed
Registry-backed discovery
Search public MCP servers through the official Registry API, with local caching and a built-in offline starter set.
Resource-aware operation
Expose available and installed capability state as MCP resources so the model can inspect the stack before changing it.
Prompt-aware composition
Ship reusable prompts for composing lean agent stacks and reviewing MCP configuration risk.
Safer configuration
Back up config before edits, avoid inventing secrets, and register remote servers without unnecessary local installs.
Current Surface
| Layer | What describe exposes |
|---|---|
| Tools | list, search, install, uninstall, installed, config-add, config-remove, config-list, config-backup, config-restore, registry-refresh |
| Resources | describe://registry/available, describe://servers/installed, describe://guide/agent-stack |
| Prompts | compose-agent-stack, harden-mcp-config |
From The Lab
KeppyLab builds small, sharp AI systems: developer tools, research infrastructure, agent evaluation workflows, and experiments that make software easier to operate through language.
WorldEnder.ai - RAG-powered text adventures with coherent long-horizon world state.
Disease Lab - Knowledge graph AI for rare disease literature and discovery workflows.
cobol-reporter - RAG and report generation for understanding COBOL systems.