MCP server implementing the cq SDK Store SPI on top of a git working tree. Provides a git-backed knowledge store with OKF markdown documents and append-only event files for confirmations and flags.
Five tools with complete input schemas and descriptions, but significant quality gaps. Tool names lack action verbs (query, propose, confirm, flag, status are weak; should be search_units, create_unit, confirm_unit, flag_unit, get_status). Descriptions are present but generic (avg ~100 chars, below 194-char baseline). Parameters lack type constraints (no enums for 'reason' in flag tool, no min/max for 'limit' in query). Output schemas are not documented, LLMs cannot plan downstream calls. Error handling is absent from descriptions. STDIO transport caps protocol readiness at 50, which cascades to overall score.
Confirms an existing knowledge unit by incrementing its confirmation count and boosting confidence. Creates an append-only confirmation event file.
Flags an existing knowledge unit with a reason (e.g., stale, duplicate). Creates an append-only flag event file and reduces confidence.
Stores a new knowledge unit as a markdown document. The unit's Evidence and Flags become the immutable seed.
Returns units sharing at least one domain with the query, ranked by the SDK's shared relevance*confidence scorer and truncated to the limit.
Aggregates store statistics including total unit count and tier distribution.
Tool names lack action verbs. 'query', 'propose', 'confirm', 'flag', 'status' are weak; should be 'search_units', 'create_unit', 'confirm_unit', 'flag_unit', 'get_status'. LLMs infer intent from names, weak names cause wrong tool selection.
No output schemas documented. LLMs cannot plan downstream calls or extract required fields. E.g., does 'query' return unit_id, domains, confidence? Does 'propose' return the new unit_id for chaining to confirm/flag?
Parameter constraints missing. 'limit' in query has no min/max (baseline: 1 - 100). 'reason' in flag has no enum (should be: stale|duplicate|incorrect|superseded). Free-form strings invite hallucinated values.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 44 | 2026-07-28+ | v2 |
Descriptions are generic and lack context. 'Returns units sharing at least one domain' (query) does not explain WHEN to use it vs other discovery tools, or what 'shared relevance*confidence scorer' means to an LLM. Avg 100 chars vs 194-char baseline.
No error handling guidance. If 'confirm' is called with a non-existent unit_id, what happens? Can the LLM retry? Should it search first? Descriptions must include recovery hints.