easy-agent-mem
Server Configuration
Describes the environment variables required to run the server.
| Name | Required | Description | Default |
|---|---|---|---|
| AGENT_MEM_STORAGE_DIR | No | Use a different local directory for structured memory. |
Instructions
Guidance the server publishes about itself, which clients place ahead of the tool catalog so the model reads it before choosing anything.
This server publishes no instructions, or was last inspected before Glama recorded them.
Capabilities
Features and capabilities supported by this server
Protocol revision2025-11-25
| Capability | Details |
|---|---|
| tools | {
"listChanged": false
} |
| prompts | {
"listChanged": false
} |
| resources | {
"subscribe": false,
"listChanged": false
} |
| experimental | {} |
Tools
Functions exposed to the LLM to take actions
| Name | Description |
|---|---|
| query_memoryB | Load saved memory before work. Explicit session IDs enable durable scoped continuity. |
| summarize_to_obsidianC | Save a durable summary. MCP/explicit sessions are isolated; direct legacy calls keep Markdown behavior. |
| list_recent_sessionsC | List legacy notes and summaries visible to this caller; never other private sessions. |
| session_statusC | Check the memory-loaded state of this session only. |
| get_task_contextC | Get whole-record context with citations, constraints, freshness and explicit budget overflow. |
| record_decisionC | Save a sourced decision/constraint/blocker. Sharing and supersession are explicit. |
| check_decisionsC | Flag changed or missing evidence without rewriting decisions. |
| record_review_outcomeC | Record review/failure evidence. Commands are stored data; verification is caller-reported. |
| get_review_memoryC | Retrieve findings, dismissals, fixes and failed approaches without promoting them to proof. |
| share_memory_recordC | Explicitly promote a visible private record; no automatic promotion of agent claims. |
| get_session_handoffC | Return the durable session handle needed to resume private memory in a new connection. |
Prompts
Interactive templates invoked by user choice
| Name | Description |
|---|---|
No prompts | |
Resources
Contextual data attached and managed by the client
| Name | Description |
|---|---|
No resources | |
TDQS
Scored across 11 tools
Most tools have clearly distinct scopes: retrieval (query_memory, get_task_context, get_review_memory, get_session_handoff), writing (record_decision, record_review_outcome, summarize_to_obsidian), and session inspection (session_status, list_recent_sessions). There is some adjacency among the three retrieval tools and between session_status and get_session_handoff, but the descriptions give enough signals (citations/budget vs. raw memory vs. review findings vs. session handle) to pick correctly.
Predominantly consistent verb_noun snake_case (record_decision, query_memory, get_review_memory, check_decisions, list_recent_sessions). Minor deviations exist: session_status is a bare noun phrase with no verb, and summarize_to_obsidian adds a preposition plus a backend-specific name, which is less predictable than the rest.
Eleven tools is well within a healthy range and each maps to a concrete memory-lifecycle action (record, query, verify, share, summarize, handoff, inspect session). No redundant near-duplicates bloat the surface.
Core read/write paths for decisions, review evidence, sharing, summaries, and session handoff are covered. However, there is no explicit delete/forget/prune or supersede operation for a memory system where stale or private data must be removed, and check_decisions flags problems without any tool to resolve them, leaving dead ends.