Command-line interface for Dakera AI Agent Memory Platform — manage AI agent memory, sessions, and knowledge graphs
Dakera CLI exposes 53 tools via STDIO MCP with consistent naming (verb_noun pattern) and basic descriptions. However, the implementation suffers from several critical gaps: (1) Parameter descriptions are often minimal or missing context (e.g., 'memory_type' tools lack enum definitions for the 4 memory types); (2) Output schemas are not documented, callers cannot predict response structure; (3) Error handling provides no recovery guidance; (4) Tool definitions are inferred from Rust CLI argument builders rather than explicitly registered with full JSON Schema; (5) Many parameters lack type definitions in visible schema (e.g., namespace_delete 'yes' and 'dry_run' shown as bool but no 'required' annotation). Naming is solid (all start with verbs, follow verb_noun pattern), but parameter documentation and output contracts fall short of production standards. The server lands in the 'fair to poor' range due to missing schemas and limited descriptions.
Create a backup
Delete a backup
List backups
Restore from a backup
Clear cache for a namespace or all namespaces
Get cache statistics
Get cluster nodes information
Get cluster status
Get server configuration
Output schemas completely undocumented. No MCP tool registration shows response structure, LLMs cannot plan downstream calls or parse results correctly.
Enum constraints missing for bounded parameters. 'memory_type' accepts 4 specific values (working|episodic|semantic|procedural) but is defined as free-form string, inviting hallucinated values.
Decay curves ('exponential|power_law|logarithmic|flat') in namespace_policy_set lack enum constraints, presented as free-form strings.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 46 | 2026-07-28+ | v2 |
Update server configuration
Configure TTL settings for a namespace
Get index statistics for a namespace
Optimize a namespace (compact storage and indexes)
Get namespace quotas
Update namespace quotas
Trigger index rebuild for a namespace
Get slow queries
List all agents
Get memories for an agent
List sessions for an agent
Get agent statistics and memory breakdown
Generate bash shell completion script
Generate fish shell completion script
Generate zsh shell completion script
Add or update a named server profile
List all profiles
Switch the active profile
Show current configuration
Check server health and connectivity
Get full-text index statistics
Rebuild index
Get index statistics
Interactive setup wizard — configure server URL and default namespace
Trigger memory consolidation
Provide feedback on memory quality
Delete a memory
Get a specific memory by ID
Update memory importance scores
Recall memories for an agent using similarity search
Full-text search across memories
Store a memory for an agent
Update a memory
Create a new namespace
Delete a namespace and all its data
Get namespace information
List all namespaces
Show the current memory policy for a namespace
Update memory policy fields for a namespace (only supplied flags are changed)
End an active session
Get session information
List sessions for an agent
Get memories from a session
Start a new session
No error recovery guidance. Tools like namespace_delete, memory_forget, and admin_backup_restore are destructive but provide no guidance on what to do if the operation fails or what preconditions must hold.
Confirmation/dry-run patterns exist in code (namespace_delete, index_rebuild accept 'yes' and 'dry_run' flags) but are not documented in tool descriptions, LLMs won't discover them.
Tool definitions are inferred from Rust CLI argument parsing rather than explicitly registered as JSON Schema in MCP server code, cannot verify exact parameter types and constraints.
Descriptions are generic and brief (avg ~50 chars). Many lack context on WHEN to use the tool or what happens on success/failure. 'List all namespaces' could mention how pagination works or if results are empty.
No pagination documented in list/search tools. memory_search and similar tools accept 'top_k' but don't document max result caps, total counts, or cursor handling.
Parameter 'importance' (memory_store, memory_update, memory_importance) lacks range/format documentation. Accepts f32 but no min/max bounds stated, LLMs may pass invalid values like 10.0 or -5.0.
Config 'data' parameter in admin_quotas_set is vague ('JSON data for quotas'). LLMs cannot infer the expected JSON structure without a schema example.