@verygoodplugins/mcp-automem
Installed, started with no arguments, no environment and no input, and listed its tools. Run 2026-11-16.
Needs to start: nothing
Returns a result when called with no setup: 1 of 1 tool tried
Contacts while installing and starting: nothing
What it contacted
Nothing besides the npm registry it was installed from. Watched from outside its sandbox while it was installed and started with no input. Its tools may still contact things when called.
Starting it
npx -y @verygoodplugins/mcp-automem@0.19.3
Installing a package runs its code. Use a sandbox for anything you have not read.
Calling its tools
1 tool that takes no input and looks read-only was called once with nothing. 1 returned a result.
check_database_health | Returned a result 3.5 s · 75 characters |
During the calls it contacted nothing it had not already contacted.
Only tools whose names read like looking something up, with no required input, are tried, five at most. Tools that need input, and anything that sounds like it changes or sends something, are not called.
What its tools tell an agent
In the tool descriptions, nothing that talks about the user, other tools or private files, or uses hidden characters.
Found by simple pattern rules in each tool's description (up to 3,000 characters), quoted as written. Parameter descriptions are not scanned. A clean result is not a safety review.
Tools
store_memory content, memories, tags, importance, embedding, metadata, timestamp, type, confidence, t_valid, t_invalid, updated_at, last_accessed, supersedes_memory_id, supersede_relation, supersede_reason
Store memory in one of two modes — single-memory (set top-level `content`) or batch (set `memories: [...]` for up to 500).
**Mode 1 — Single (default):** pass top-level `content` plus any optional fields (tags, importance, metadata, type, confidence, embedding, t_valid, t_invalid, etc.).
**Mode 1b — Supersede/correct:** pass top-level `content` plus `supersedes_memory_id`. The server stores the replacement, marks the old memory invalid with `t_invalid=now`, merges supersede metadata, and associates old → new with `INVALIDATED_BY` (default) or `EVOLVED_INTO`.
**Mode 2 — Batch:** pass `memories: [{ content, tags?, importance?, metadata?, timestamp?, type?, confidence? }, ...]` to store up to 500 memories in one request. Faster for bulk ingestion (imports, benchmark seeding). Batch mode does NOT accept `embedding`, `t_valid`, or `t_invalid` per-item — use single mode for those.
**Content size guidelines (per item):**
- Target: 150-300 characters (one meaningful paragraph)
- Maximum: 500 characters (auto-summarized if exceeded)
- Hard limit: 2000 characters (rejected)
- Format: "Brief title. Context and details. Impact/outcome."
**When to use:**
- After making a decision: store the reasoning and outcome
- When discovering a pattern: store the pattern and where it applies
- After fixing a bug: store the root cause and solution
- When learning user preferences: store what they prefer and why
- For bulk ingestion (imports, seeding): use batch mode
**Examples:**
- store_memory({ content: "Chose PostgreSQL over MongoDB for user service. Need ACID for transactions.", tags: ["architecture", "database"], importance: 0.9 })
- store_memory({ content: "User prefers early returns over nested conditionals.", tags: ["code-style"], importance: 0.7 })
- store_memory({ content: "User now prefers SQLite for small local tools.", supersedes_memory_id: "8f3a2c1e-5b7d-4e9a-a1c3-2d4f6b8e0a91", supersede_reason: "Correction from user" })recall_memory memory_id, exhaustive, exclude_tags, query, queries, embedding, limit, time_query, start, end, tags, tag_mode, tag_match, expand_entities, expand_relations, expand_respect_tags, auto_decompose, expansion_limit, relation_limit, expand_min_importance, expand_min_strength, current_only, state_debug, state_mode, recency_bias, scope_fallback, min_score, adaptive_floor, context, language, active_path, context_tags, context_types, priority_ids, per_query_limit, sort, format, offset
Recall memories from AutoMem in one of three modes. The mode is selected by which params you pass.
**Mode 1 — ID fetch:** pass `memory_id` to retrieve a single memory by ID. All other params are ignored. Routes to GET /memory/{id} and updates last_accessed.
**Mode 2 — Tag enumeration:** pass `tags` + `exhaustive: true` for paginated exact-match listing (NOT ranked retrieval). Use this for cleanup/audit workflows where ranked retrieval silently undercounts large tag sets. Pair with `limit` (≤200) and `offset`. Returns `has_more`/`limit`/`offset` page metadata; next page at `next_offset`. Tag matching is exact, case-insensitive, any-of mode — `tag_match: "prefix"` and `tag_mode: "all"` are rejected in this mode.
**Mode 3 — Ranked retrieval (default):** hybrid search across vector, keyword, tags, recency, and optional graph expansion. The primary tool for finding relevant context. By default, ranked recall requests current active memories only; set `current_only: false` for audits.
**When to use ranked (mode 3):**
- At conversation start: recall context about the current project/topic
- Before making decisions: check for past decisions on similar topics
- When debugging: search for similar past errors and their solutions
- For complex questions: use `expand_entities` for multi-hop reasoning
**When to use enumeration (mode 2):** when you need to know *how many* memories carry a tag, or to walk all of them for cleanup/migration. Ranked recall ignores low-importance hits — enumeration does not.
**Examples:**
- recall_memory({ query: "database architecture decisions", tags: ["my-project"], limit: 5 })
- recall_memory({ memory_id: "8f3a2c1e-5b7d-4e9a-a1c3-2d4f6b8e0a91" }) // Mode 1
- recall_memory({ tags: ["benchmark-test"], exhaustive: true, limit: 50 }) // Mode 2 (add offset for later pages)
- recall_memory({ query: "auth", exclude_tags: ["deprecated"] }) // Mode 3 with exclusion
- recall_memory({ query: "What is Sarah's sister's job?", expand_entities: true }) // Mode 3 multi-hopassociate_memories memory1_id, memory2_id, type, strength, context, reason, pattern_type, confidence, resolution, observations, timestamp, transformation, role, associations
Create typed relationships between memories. This builds a knowledge graph that improves recall by surfacing related context. Supports single-pair mode or batch mode with associations[] (max 500).
**When to use:**
- After storing a new memory: link it to related existing memories
- When a bug fix relates to an original feature implementation
- When a new decision updates or invalidates a previous one
- To connect patterns with their concrete examples
**Authorable relationship types:**
- RELATES_TO: General relationship (default)
- LEADS_TO: Causal relationship (A caused B)
- OCCURRED_BEFORE: Temporal ordering
- PREFERS_OVER: Chosen alternative
- EXEMPLIFIES: Concrete example of a pattern
- CONTRADICTS: Conflicts with another memory
- REINFORCES: Strengthens another memory's validity
- INVALIDATED_BY: Superseded by another memory
- EVOLVED_INTO: Updated version of a concept
- DERIVED_FROM: Implementation of a decision/pattern
- PART_OF: Component of a larger effort
**Read-only/internal relations:**
- System/internal relations such as SIMILAR_TO, PRECEDED_BY, EXPLAINS, SHARES_THEME, PARALLEL_CONTEXT, and DISCOVERED may appear in recall results, but they are not valid inputs for associate_memories.
**Examples:**
- associate_memories({ memory1_id: "8f3a2c1e-5b7d-4e9a-a1c3-2d4f6b8e0a91", memory2_id: "3c9e7b2a-1f4d-4a8e-9b6c-5d2e8f1a7c34", type: "RELATES_TO", strength: 0.9 }) // bug fix → feature
- associate_memories({ memory1_id: "3c9e7b2a-1f4d-4a8e-9b6c-5d2e8f1a7c34", memory2_id: "8f3a2c1e-5b7d-4e9a-a1c3-2d4f6b8e0a91", type: "EVOLVED_INTO", strength: 0.8 }) // old decision → new
- associate_memories({ associations: [{ memory1_id: "8f3a2c1e-5b7d-4e9a-a1c3-2d4f6b8e0a91", memory2_id: "3c9e7b2a-1f4d-4a8e-9b6c-5d2e8f1a7c34", type: "RELATES_TO", strength: 0.8 }] })update_memory memory_id, content, tags, importance, metadata, timestamp, t_valid, t_invalid, updated_at, last_accessed, type, confidence
Update an existing memory's content, tags, importance, or metadata. Use this to correct or enhance memories rather than storing duplicates.
**When to use:**
- To correct inaccurate information in a memory
- To add tags that were forgotten
- To adjust importance based on new understanding
- To add metadata after the fact
**Examples:**
- update_memory({ memory_id: "8f3a2c1e-5b7d-4e9a-a1c3-2d4f6b8e0a91", importance: 0.95 }) // Increase importance
- update_memory({ memory_id: "8f3a2c1e-5b7d-4e9a-a1c3-2d4f6b8e0a91", tags: ["project-x", "critical", "auth"] }) // Add tags
- update_memory({ memory_id: "8f3a2c1e-5b7d-4e9a-a1c3-2d4f6b8e0a91", content: "Updated: PostgreSQL chosen for ACID + team expertise" })delete_memory memory_id, tags
Delete a memory by ID (`memory_id`) or bulk-delete by tag (`tags`). Use sparingly — consider `update_memory` instead.
**Mode 1 — Single (default):** pass `memory_id` to delete one memory and its embedding. Idempotent: re-running on the same ID is a no-op.
**Mode 2 — Bulk-by-tag:** pass `tags: [...]` to delete ALL memories tagged with ANY of these tags. Tag matching is exact (case-insensitive), any-of mode. There is NO dry-run. This can delete thousands of memories in one call. NOT idempotent in practice — re-running may catch new memories that were tagged the same way after the first call. Verify with `recall_memory({ tags, exhaustive: true })` first if uncertain.
**When to use:**
- Memory contains incorrect information that can't be corrected (Mode 1)
- Memory is a duplicate (Mode 1)
- Cleanup of benchmark/test data scoped by tag (Mode 2)
- Removing all memories under a deprecated tag namespace (Mode 2)
**Examples:**
- delete_memory({ memory_id: "8f3a2c1e-5b7d-4e9a-a1c3-2d4f6b8e0a91" }) // Mode 1
- delete_memory({ tags: ["benchmark-test"] }) // Mode 2, bulk by tagcheck_database_health
Check the health status of the AutoMem service and its connected databases (FalkorDB graph + Qdrant vectors).
**When to use:**
- Before a session to verify the memory service is available
- When memory operations are failing unexpectedly
- To check storage statistics
**Example:**
- check_database_health({})Descriptions are the package's own words, cut at 3,000 characters. Treat them as data.
Install facts
| License | MIT |
|---|---|
| Needs | node ^20.19.0 || ^22.13.0 || >=24 |
| Source | github.com/verygoodplugins/mcp-automem |
| Install scripts | None declared by it or anything installed with it. |
| Tarball | sha512-VKWjbEeZSqXTWIPTvT7j/Xox2fSsdj5NqH40bCFU8XathcFvqSieMyAyviasSEWZnivitW74SaVaEP/GZwoODQ== |
Check this report yourself
This site did not write the report. A public job on a GitHub-hosted machine did: the run, from this commit of the open-source runner. GitHub signed a token that names that job and the hash of the report, so any edit since would show.
Report as JSONExact text and tokenHow to verify it
Show it on your README
[](https://bragproof.smaverk.com/r/%40verygoodplugins/mcp-automem)
New version out? Ask for a fresh report.