Context vault for coding agents - stores and retrieves feature memories, discussions, Jira integrations, attachments, and git branch mappings with full-text search and fuzzy matching capabilities
MemoryVault MCP presents a moderately well-defined tool set with solid parameter schemas and mostly clear descriptions. All 7 tools have explicit input schemas via Pydantic models and descriptive text. However, there are significant gaps in output schema documentation (no documented return types visible), error handling guidance is missing entirely, and several tools exhibit overly broad responsibilities that violate single-concern design. The schema quality is above average (most params have type+description), but the lack of output documentation and error recovery patterns pushes the score into the 'Fair' range. Tool naming is clear and verb-driven, descriptions are adequate (100-250 chars range, within baseline), but composition issues (e.g., get_feature_memory_context doing too much) and missing error guidance prevent a higher score.
Get implementation-ready guidance extracted from discussions: requirements, edge cases, changes, conflicts, clarifications, and unresolved items.
Get chronological timeline of discussions with optional filters by decision type (requirement, edge-case, change, conflict, clarification) and date range.
Get complete context packet for a feature memory including discussions, attachments, branches, and Jira info. Returns normalized JSON for AI agents.
Check MCP server health including database connection, version, and settings.
List git branches related to a feature memory with repository information.
Resolve feature memory by Jira story key, ID, or name. Priority: exact story key → story ID → feature ID → fuzzy title match. Returns matching feature memories with confidence scores.
Output schemas completely undocumented across all tools. LLMs cannot plan downstream tool calls or extract required fields without knowing response structure.
get_feature_memory_context violates single-responsibility principle by gathering discussions, attachments, branches, AND Jira info in one call. Should decompose into separate tools (or keep as a convenience wrapper but provide fine-grained alternatives).
No error recovery guidance across any tool. Missing documentation of what errors can occur, whether they are retryable, and what the LLM should do next (e.g., 'Feature memory not found. Try search_discussions to find related content first.').
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | C | 63 | <=2025-11-25 | v2 |
| 2026-03-09 | D | 57 | - | v1 |
Search discussions across feature memories with text query and filters (story key, decision type, tags). Returns relevant discussion snippets.
Pagination not addressed. search_discussions accepts a limit param but no offset/page/cursor fields documented. What happens when there are 1000+ results? No guidance on whether results are truncated or paginated.
list_related_branches has minimal description (10 chars) below 20-char baseline. Insufficient context for LLM to understand when to call it or what 'repository information' means.
No tool annotations (readOnlyHint, idempotentHint) despite all tools being read-only operations. Modern MCP servers should declare safety properties.