René's URL Explorer Experiment

{"@context":"https://schema.org","@type":"DiscussionForumPosting","headline":"[FEATURE] Consume the naked upstream HTTP status code from any-sdk and remove response-text munging","articleBody":"## Description\n\nThe issue #670 fix (MCP tools surface upstream HTTP errors instead of empty result sets, with 404 exempted as database-semantics \"zero rows\") currently recovers the upstream status by parsing any-sdk's error message text, because the status code is not carried through the any-sdk invoker boundary (`providerinvoker.Result` is just `{Body, Messages}`):\n\n- `internal/stackql/execution` (`isUpstreamNotFound`): the structured fragment regex `(?i)status code:?\\s*(\\d{3})` is authoritative when present (a 403 whose body mentions \"not found\" is still an error); unstructured bodies (HTML / XML / plain text published verbatim) fall back to a loose, any-case match:\n\n~~~go\nvar upstreamNotFoundLooseRegex = regexp.MustCompile(`(?i)\\b404\\b|not\\s+found`)\n~~~\n\n- `internal/stackql/mcpbackend` (`classifyBackendError`): the same fragment regex classifies MCP tool errors as `upstream http error: {\"http_status\": N, \"retryable\": bool}` (408/429/5xx retryable).\n\nThis works for mainstream provider error shapes but is fragile by construction: the status is only present when any-sdk happens to format it into the text, and the loose fallback trades precision for safety (a false positive restores the historical empty-result behaviour, which is the safe direction).\n\n## Proposal\n\nOnce \u003clink to any-sdk issue\u003e lands and the any-sdk dependency is bumped:\n\n1. Read `StatusCode` from the invoker result in `internal/stackql/execution` (`monoValentExecution.GetExecutor`): exempt 404 by value, error on other 4xx/5xx under strict upstream error semantics (MCP only).\n2. Classify by value in `internal/stackql/mcpbackend` (`classifyBackendError`), keeping the wire contract (`http_status`, `retryable`) unchanged.\n3. Delete `upstreamStatusCodeRegex` (both copies), `upstreamNotFoundLooseRegex` and the text handling in `isUpstreamNotFound`.\n\nThe behaviour contract is already pinned by robot scenarios in `test/robot/functional/mcp.robot` and stays unchanged as the regression net:\n\n~~~text\nMCP HTTP Upstream 403 Surfaces As Non Retryable Tool Error\nMCP HTTP Upstream 429 Surfaces As Retryable Tool Error\nMCP HTTP Upstream 404 Returns Empty Result Set\n~~~\n\n## Depends on\n\n[\u003clink to any-sdk issue\u003e](https://github.com/stackql/any-sdk/issues/115), plus the next any-sdk release/bump.\n","author":{"url":"https://github.com/jeffreyaven","@type":"Person","name":"jeffreyaven"},"datePublished":"2026-07-03T09:50:17.000Z","interactionStatistic":{"@type":"InteractionCounter","interactionType":"https://schema.org/CommentAction","userInteractionCount":0},"url":"https://github.com/681/stackql/issues/681"}