What changed
| Before | Now |
|---|---|
| Install Python and start an MCP server | Run one npx installer |
| Connect an MCP transport | Your agent discovers local skills |
| Call MCP tools or HTTP endpoints | Ask naturally or run a bundled script |
| Python runtime and optional integrations | Node.js 20 and self-contained ESM runners |
| Server-formatted responses | One consistent JSON envelope |
Natural-language replacement
Instead of connecting a server and selecting a remote tool, ask the agent for the analysis. The installed skill activates from its description, reads only the references it needs, and executes locally.
Estimate density altitude at 5,000 ft pressure altitude, 30 C, and 10 C dew point. Include units and operational limitations.Unsupported legacy interfaces
The following interfaces are intentionally removed: MCP, REST and FastAPI, Docker and Kubernetes deployment, the old public Python CLI, and direct Python imports. Git history preserves the previous implementation for maintainers who need to inspect it.
Operation names
The 45 retained calculation names remain recognizable. Discovery and LLM formatting helpers were removed because skill activation and the runner envelope now provide those responsibilities.
Do not reuse former Python signatures
The public operation names are retained, but the supported interface is now a JSON object passed to the local runner. Inspect the installed skill's example and operation reference instead of assuming a former Python function signature is unchanged.
Need exact control?
Use the runner contract directly inside any installed skill:
node scripts/run.mjs <operation> --input-json '<json>'Successful output includes the result, units, assumptions, warnings, and provenance. Invalid input exits nonzero; it is not silently repaired with inferred safety-relevant values.