Engram

Install

Works with Claude Code, Codex, VS Code, Cursor, Windsurf, Claude Desktop. Read the repository before you give any tool access to your data.

npx agentmarket install connector/devinmlowe-engram

About

No reviews yet. Installed it? Reviews come from verified installers only. How reviews work

Permissions

What this server can do and what it asks for, from its tool list and install config. Checked Oct 25, 2026.

Tools

9 read · 6 write · 1 delete · 1 not yet classified

Secrets

1 (API keys, tokens or passwords)

Install scripts

Not scanned yet

Runs as

Node.js package

Tools (17)

Read from the source code; the running server may differ. Descriptions are the server’s own words; hidden characters are shown as [U+…].

• commitmentsCommitmentsread-only

  List tracked commitments — first-person promises, intentions and follow-ups owed by others, extracted nightly from conversations ("mention once, never dropped"). Default: pending items, overdue first, then by due date, then newest. Use include_due_within_days to surface only what is due soon or overdue.

• commitments_updateUpdate Commitment

  Resolve a tracked commitment once the user confirms it is handled: mark it done, dropped, or superseded by another commitment. Accepts the full id or a unique prefix (6+ characters).

• exploreExplore Knowledge Graphread-only

  Explore connections in the knowledge graph starting from an entity. Shows what a concept, tool, project, or technology is connected to. Use after recall to understand how things relate to each other.

• explore_selectiveSelective Graph Explorationread-only

  Explore the knowledge graph selectively by expanding only branches relevant to specified criteria. Unlike explore (fixed-depth BFS), this uses embedding similarity to prune irrelevant neighbors and recursively follows only relevant paths. Use when you want to find connections related to a specific topic or question.

• extract_commitments

  Record the commitments, intentions and follow-ups actually stated in the conversation. Return an empty list when nothing qualifies.

• fetch_snippetsFetch Snippetsread-only

  Fetch multiple line ranges from a single file in one call. Ranges are merged when overlapping, padded with optional context lines, and joined with gap markers. More efficient than multiple show calls for targeted code reading.

• forgetForget Memorycan change or delete data

  Remove a memory the user says is wrong or stale. Pass the memory_id from a recall result (<semantic id="…">) to forget it in one call. Pass query instead to search for candidates: the tool returns matching memories with their ids and forgets nothing unless confirm is true AND exactly one memory matches (a single result, or a single result whose content equals the query). The memory is soft-deleted (kept for ENGRAM_FORGET_RETENTION_DAYS, then purged by the dream pipeline), drops out of every recall path immediately, is logged in the change log with this client's name, and will not be re-extract…

• index_file_structureIndex File Structure

  Parse a source file to extract function, class, and module definitions, then index them as entities in the knowledge graph with 'contains' relationships. Enables RLM to discover file contents without reading the full file.

• ingest_turnIngest Turn

  Record one user/assistant turn of an external agent conversation (e.g. a Hermes profile) in engram's episodic layer so it becomes searchable and feeds the nightly dream extraction. Idempotent: re-sending the same session_id + turn_index updates the turn in place. The conversation carries the given tenant scope, and every memory later extracted from it inherits that scope. An optional author {id, name, is_bot} is stored on the turn.

• recallRecall Memoriesread-only

  Retrieve relevant memories from past Claude Code conversations and extracted knowledge. Uses hybrid semantic + keyword search with token-budgeted output across episodic and semantic memory stores. Search BEFORE every task to recover decisions, solutions, and context. Time-scope with after/before (YYYY-MM-DD) or a natural-language dateHint ("last week", "in March", "this day last year", "on this day"); dateBasis picks whether dates mean when a memory was filed or when the events it describes happened. The applied window is echoed in <date_filter>. Retrieval records access bookkeeping on the ret…

• recall_drillDrill Into Resultread-only

  Drill into a specific result from a recall session to get expanded context. For episodic results: shows surrounding conversation exchanges. For semantic results: shows source conversation segments. For graph results: shows entity with full relationship neighborhood. Retrieval records access bookkeeping on the returned memories (access_count, last_accessed, stability); pass reinforce: false to opt out.

• recall_sessionRecall Sessionread-only

  Create or continue an iterative search session for multi-step memory exploration. Omit session_id to start a new session; provide session_id to refine with a new query. Sessions track accumulated results and remaining token budget. Use recall_drill to expand individual results. Retrieval records access bookkeeping on the returned memories (access_count, last_accessed, stability); pass reinforce: false to opt out.

• reflectReflect on Knowledge Graph

  View emergent patterns and structure in the knowledge graph. Shows topic communities with meaningful names, bridge entities connecting different domains, temporal patterns, and graph health. Use after working on a topic to understand how it connects to other knowledge domains.

• rememberRemember

  Store a fact, preference, decision, or other knowledge as a semantic memory. Use this to explicitly record important information that should persist across conversations. Automatically deduplicates against existing memories.

• remember_batchBatch Remember

  Store multiple facts, preferences, decisions, or other knowledge items as semantic memories in a single call. Supports up to 50 items per batch. Automatically deduplicates against existing memories and within the batch. Use for bulk ingestion from RLM agents or dream pipeline. Optionally link each memory to existing graph entities via relates_to_entities.

Signals

Stars

1

Forks

0

Open issues

33

Weekly downloads

30

Last commit

Sep 21, 2026 (1 month ago)

Last release

Sep 21, 2026 (1 month ago)

Last package publish

Sep 21, 2026 (1 month ago)

From GitHub (fetched Oct 24, 2026) and the package registry.

Details

Registry name

io.github.devinmlowe/engram

Version

0.5.0

Auth

API key

Packages

npm @devinmlowe/engram

License

MIT

Repository

github.com/devinmlowe/engram

Website

github.com/devinmlowe/engram#readme

First seen

Oct 24, 2026

Last checked

Oct 25, 2026

Source

Official MCP Registrygithub-graphql