Loading ToolBench

Reading the current corpus…

claims-ledger MCP server · D · ToolBench

claims-ledger

MCP server for Evidence Decision Trace — anchors claims and decisions to verifiable sources, fails CI when anchors go stale

2026-07-28+STDIO5 tools

54/100

Definition Quality76

Protocol Readiness35

Supportability29

Spec posture

Inferred effective spec 2026-07-28+·current baseline 2026-07-28

Static source inference · medium confidence

Current-spec patterns detected

Summary

Claims-Ledger provides 5 domain-specific tools with clear, detailed descriptions (avg 150 chars) and well-structured input schemas. Tool names follow verb_noun convention (ledger_search, trace_add_decision, anchor_verify). Schemas include proper types, enums, and constraints. However, output schemas are not documented in the visible code, and error handling guidance is minimal. The domain is specialized (source-anchored claims verification) but lacks recovery hints for common failure modes.

Tools (5)

anchor_verifyread onlysource verified75/100

Freshness-check a single anchor ref (optionally with its quote) against the working tree.

ledger_claimsread onlysource verified75/100

List existing claims (optionally filtered by scope or path substring) so an agent knows which claims its diff might invalidate.

ledger_searchread onlysource verified80/100

Keyword search over the repo's Claims Ledger and the open trace — find anchored evidence BEFORE deciding.

trace_add_decisionwritesource verified83/100

Append an anchored decision to the open trace. Each anchor's quote MUST literally resolve at its ref — fabricated quotes are rejected.

trace_scorewriteauthsource verified68/100

Run the cross-family judge panel over every decision⟷evidence pair in the open trace (median + spread>20 flagging). Skips gracefully without OPENROUTER_API_KEY.

Issues (5)

HIGHtool ↗

Output schemas not documented. LLMs cannot infer what ledger_search, trace_score, or anchor_verify return, forcing them to guess field names and types for downstream reasoning.

ledger_searchtrace_scoreanchor_verifyledger_claims

HIGHtool-description ↗

trace_score has empty input schema ({}). No parameters documented. Description mentions OPENROUTER_API_KEY but does not explain how to provide it or what happens if missing.

trace_score

MEDIUMrecovery-guide ↗

Error handling lacks recovery guidance. No tool description explains what to do if an anchor quote fails to resolve, if the ledger is malformed, or if the OpenRouter API is unavailable.

trace_add_decisiontrace_scoreanchor_verify

MEDIUMpaginated-result ↗

ledger_search and ledger_claims lack pagination guidance. No mention of result limits, total counts, or how to handle large ledgers. Default k=5 is reasonable but undocumented.

Recommendations

Document output schemas for all tools. Example: ledger_search should return {results: [{id, path, quote, confidence}], total_count, next_cursor}. anchor_verify should return {valid: boolean, resolved_quote: string, error?: string}.
Add error recovery hints to descriptions. E.g., trace_add_decision: 'If quote fails to resolve, verify the ref URI format (git://sha/path#L1-L2) and that the quote is verbatim. Call anchor_verify first to test.'
Document trace_score's dependency on OPENROUTER_API_KEY. Clarify: 'Requires OPENROUTER_API_KEY env var. Skips gracefully if missing, returning empty results. Call this after adding decisions to validate cross-family consistency.'
Add pagination to ledger_search and ledger_claims. Include limit (default 20, max 100), offset/cursor, and total_count in responses. Document: 'Results capped at 100 per call. Use offset to iterate large ledgers.'
Specify what 'confidence' enum values (high/medium/low) mean in trace_add_decision anchors. E.g., 'high: quote is exact and unambiguous; medium: quote is paraphrased or context-dependent; low: quote is inferred or approximate.'

Score history

Overall score trend

First recorded score · v2 rubric

54/100

Scored	Grade	Overall	Spec posture	Rubric
2026-09-23	D	54	2026-07-28+	v2