Git-backed long-term memory authority for AI agents via MCP. Provides project-scoped long-term memory and cross-conversation continuity with validated Git-backed Authority.
MemAuthority has 16 tools with reasonable naming (verb_noun pattern) and descriptions present for all. However, critical gaps undermine quality: most tools lack visible input schemas in the source code, only RouteArgs is explicitly defined. Parameter descriptions exist in struct tags but are not formally validated against JSON Schema. Output schemas are entirely undocumented. Error handling is absent from tool definitions. The server conflates read-only and destructive operations without clear guidance on idempotency or retry safety. STDIO transport caps protocol readiness at 50, further limiting production viability.
Append progress entry to project memory
Archive a project in memory authority
Create a new project in memory authority
Mark project memory items as verified
Migrate project lifecycle state
Query declarative runtime metadata describing stored topology
Read durable memory from project resources
Input schemas not visible in source code. Only RouteArgs struct is explicitly defined; 15 other tools have parameter descriptions in comments but no formal JSON Schema validation. This violates the pattern:constrained-input requirement and makes it impossible for LLMs to validate inputs before calling.
Output schemas completely undocumented. No tool definition shows what fields are returned, their types, or structure. LLMs cannot plan downstream tool calls or extract required data without documented return types.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 46 | 2025-06-18+ | v2 |
Record a pitfall in project memory
Record a runtime observation in project metadata
Restore an archived project in memory authority
Project routing query. Matching is exact after Unicode NFC normalization and case folding against project IDs and aliases; ambiguous and archived matches are reported explicitly.
Search project memory resources
Query server status including write enablement and read-only state
Update handoff sections in project memory
Update declarative runtime metadata
Update sections in project memory roles
No error handling guidance in tool definitions. Tools marked DESTRUCTIVE (archive, restore, migrate, update) lack confirmation patterns, dry-run options, or recovery instructions. Agents cannot distinguish retryable from fatal errors.
Idempotency key parameter present but not documented. All write/destructive tools accept 'client_idempotency_key' but no description explains its purpose, format, or how it prevents duplicate operations. LLMs will not use it correctly without guidance.
Parameter descriptions lack actionable constraints. 'section_updates' and 'operations' are described as arrays but no schema shows what objects they contain, required fields, or valid values. LLMs will hallucinate structure.