The universal prompt SDK. Fetch prompts from any PLP-compliant library and use them with any AI provider.

Echostash's shared contract: zod schemas and types for prompts, MCP tool surfaces, audit reports, baselines and the server-runner eval protocol.

Read an MCP server's tools/list into a typed tool surface (stdio or HTTP), and diff it against a committed baseline.

Discovery-agnostic prompt manifest: hash prompt content, model and params, diff two manifests, and read or write baselines.

Deterministic checks and the Tool Surface Score for MCP tool definitions: confusable tools, undescribed params, thin descriptions, token budget. No model calls.

Isomorphic scorer engine for LLM evals: deterministic assertions (contains, regex, json-schema, ...) plus the scorer catalog shared by the runner and the UI.

Audit an MCP server's tool surface, scan your code for LLM prompts, and gate regressions in CI. No API key, no server, no model calls.

The universal prompt SDK. Fetch prompts from any PLP-compliant library and use them with any AI provider.