oas-mcp

Sampling Helpers

Sampling helpers expose curated subsets or transformations of prior responses / schemas to aid model prompting, summarization, and test generation.

Goals

• Avoid re-transmitting large raw responses in each prompt.
• Provide deterministic, cacheable slices.
• Support progressive enrichment (body → summary → embeddings later).

Core Helper Tools

| Tool | Purpose | | —- | ——- | | _sample_response_body | Capture and persist last response body for an operation id. | | _sample_summarize_json | Summarize captured JSON (delegates to summarize_json when available). | | _sample_list | Enumerate stored samples & metadata. |

(Names may be prefixed per collision / naming mode.)

Workflow Example

1. Call endpoint tool getUser (streaming or not).
2. Invoke _sample_response_body with { operationId: "getUser" }.
3. Invoke _sample_summarize_json to produce a compact summary resource.
4. Client fetches resource sample_response_getUser or sample_summary_getUser by name for inclusion elsewhere.

Storage & Resources

Each sample materializes as a resource with a stable name (sample_response_<op>). Hash-based change notifications allow clients to refresh selectively.

Summarization Fallback

If summarization model/tool unavailable, summary helper returns truncated JSON (size-limited) ensuring deterministic output.

Progressive Streaming Synergy

Sampling tools can operate even when original response was chunk-streamed; body assembly is performed during streaming with digest validation before storing.

Roadmap

• Embedding generation (_sample_embed_json)
• Multi-sample aggregation & schema diffing
• Token-aware prompt chunking

Next: Resources & Prompts or Streaming Responses.

This site is open source. Improve this page.