Loading ToolBench
Sefaria MCP — the free digital library of Jewish texts. Fetch any passage (Torah, Talmud, Mishnah, Midrash, commentaries) in Hebrew and/or English by reference, validate/autocomplete a reference, and list the commentaries and cross-references on a passage. Keyless.
Three well-named, read-only tools with complete input schemas and clear descriptions. All tools follow verb_noun naming (get_*, lookup_*, get_*). Descriptions are concise (60 - 180 chars) and explain purpose clearly. Input schemas include types and required fields. However, output schemas are not documented, LLMs cannot predict response structure for planning. Parameter descriptions lack format constraints (e.g., 'ref' format examples are helpful but not formal). Error handling is basic (generic error objects). No pagination support despite get_commentaries returning lists. Missing tool annotations (readOnlyHint). Overall solid for a read-only API wrapper, but lacks production-grade output documentation and error guidance.
List the commentaries and cross-references on a passage (e.g. Rashi, Targum, quoting commentaries) from Sefaria's link graph. Keyless.
Fetch a passage of Jewish text from Sefaria by reference, in Hebrew and/or English. Works for Torah/Tanakh, Talmud, Mishnah, Midrash, Halacha, Kabbalah, and commentaries. Keyless.
Validate or autocomplete a Sefaria reference or title (e.g. "Genesis", "Rashi", "Berakhot"). Returns whether it is a valid reference/book and a list of completions. Keyless.
Output schemas not documented. LLMs cannot predict response structure (fields, types, nesting) for downstream planning or error recovery.
get_commentaries accepts 'limit' parameter but no pagination cursor/offset. Returning 25 - 100 items without next_cursor or total_count violates pagination pattern.
Parameter 'language' in get_text lacks enum constraint. Description lists valid values ('en', 'he', 'both') as examples, not formal constraints. LLMs may hallucinate invalid values like 'fr' or 'es'.
Error responses are generic objects with 'error' field. No recovery guidance (e.g., 'Try lookup_ref() first to validate the reference'). Errors do not classify as retryable vs. user-fixable.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 59 | 2026-07-28+ | v2 |
Tool annotations missing. No readOnlyHint, destructiveHint, or idempotentHint declared. Clients cannot infer safety properties without explicit metadata.