engram-mcp
Server Configuration
Describes the environment variables required to run the server.
| Name | Required | Description | Default |
|---|---|---|---|
| ENGRAM_DB | No | SQLite database path | ~/.engram/memory.db |
| ENGRAM_EMBEDDING_URL | No | Ollama base URL for embeddings | http://localhost:11434 |
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 | {} |
Tools
Functions exposed to the LLM to take actions
| Name | Description |
|---|---|
| rememberB | Store a memory entry for a session. Embeddings and graph extraction happen automatically. |
| recallA | Retrieve relevant memories via semantic search (falls back to keyword). Searches working + long_term tiers. Pass userId to blend in cross-session user facts. |
| historyA | Get recent conversation history for a session in chronological order. |
| forgetA | Delete session memories. Delete all, one by ID, or entries before a date. |
| statsB | Memory statistics for a session — total, by role, by tier (working/long_term/archived), graph counts. |
| consolidateB | Consolidate old working memories into dense long-term summaries via LLM. Archives originals. |
| graphA | Query the knowledge graph for an entity — returns relationships and source memories. Requires ENGRAM_GRAPH=1. |
| remember_userA | Store a user-scoped memory that persists across all sessions. Use for preferences, identity, long-term facts. |
| recall_userB | Recall user-scoped memories — works from any session context. |
| forget_userB | Delete user-scoped memories. |
| consolidate_userB | Consolidate user-scoped working memories into long-term summaries. |
| user_statsB | Memory statistics for a user — total, by role, by tier. |
| recall_by_timeA | Query memories using natural language time expressions like "yesterday", "last week", "this morning", "before 3pm". |
| recall_recentA | Get the N most recent memories for a session — no query needed. |
| daily_summaryA | Returns a summary of memories for a given date (YYYY-MM-DD), or today if omitted. |
| temporal_statsB | Returns temporal statistics about when memories were created — frequency over time, most active periods. |
| export_memoriesB | Export all session memories as markdown or JSON. |
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 17 tools
Each tool has a clear, distinct purpose. Tools are separated by scope (session vs. user), operation (remember, recall, forget, consolidate, stats, export, graph, history), and retrieval method (semantic, time-based, recent). No overlapping functionality.
All tools follow a consistent verb_noun pattern (e.g., recall_user, forget, stats). Variations like recall_by_time, recall_recent use predictable suffixes. No mixing of naming conventions.
17 tools is appropriate for a memory management system. The number covers all necessary operations (CRUD, search, statistics, export, graph, user-scoped variants) without being excessive.
The tool set covers core memory operations: store, retrieve (multiple methods), delete, consolidate, export, and statistics. An explicit update tool is missing, but consolidation serves as an update mechanism. Minor gap, but overall comprehensive.