occam_dataset_export¶
Status: verified against src/FFOccamMcp.Core/Tools/OccamDatasetExportTool.cs @ 2026-09-29
Profiles: auditor, full — not reader (default), researcher, minimal, or basic
Contract depth: MCP API · runtime tools/list
Purpose¶
Build a signed, auditable dataset from 1–20 URLs: each URL is live-transcoded into a row; one manifest signature (when receipts enabled) binds the ordered Merkle root of row leaves. Proves integrity of the export artifact under your local key — not factual correctness.
When to use / when not¶
| Use when | Do not use when |
|---|---|
| Ordered signed corpus with per-row receipts + set-level manifest | One URL markdown → occam_transcode |
Row verify via MCP; set verify via CLI verify --mode manifest |
Bulk async without provenance → occam_batch |
| Same policy/session for every URL is enough | Per-URL token/focus knobs (not exposed here) |
Implementation: sequential rows; forced json_blocks=true, playbook_policy=auto.
Parameters¶
| Name | Required | Default | Meaning |
|---|---|---|---|
urls |
yes | — | JSON array of HTTP(S) URLs (1–20) |
backend_policy |
no | http_then_browser |
Applied to every URL |
session_profile |
no | null | Applied to every URL |
Response shape¶
Success (ok: true): always after validation — even if some rows fail. manifest (v: 1, alg: ecdsa-p256-sha256, manifestRoot, optional keyId/sig), rows[] (url, finalUrl, ok, rowLeaf, optional hashes/receipt/failureCode), timestamp.
When OCCAM_RECEIPTS is off: leaves/root remain; signatures omitted. Failed rows may get negative receipts only for selected wall codes (e.g. captcha, login, paywall, 401/403/404/410).
Tool failure: invalid_arguments only.
Related env vars¶
OCCAM_RECEIPTS, OCCAM_PROFILE — configuration.
Verification¶
- Per row:
occam_verifywithrows[i].receipt - Whole set: CLI
verify --mode manifest(not an MCP mode)