hmem — Humanlike Memory for AI Agents

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/bumblebiber-hmem-mcp

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

7 read · 11 write · 4 delete · 2 not yet classified

Secrets

None asked for

Install scripts

Not scanned yet

Runs as

Node.js package

Tools (24)

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

• append_memory

  Append new child nodes to an existing memory entry or node (your own personal memory). Existing children are preserved — new nodes are added after them. Use this to extend an existing entry with additional detail without overwriting it. Content uses tab indentation relative to the parent: 0 tabs = direct child of id 1 tab = grandchild, etc. Two modes: Simple (title+body): append_memory(id='L0003', title='New finding', body='Detailed explanation') Complex (full sub-tree): append_memory(id='L0003', content='New finding\n\tSub-detail\n\t\tDeep') Examples: append_memory(id='P0048.6',…

• create_project

  Create a new project with the standard R0009 schema. Automatically creates: 1. P-entry with all 9 L2 sections (Overview, Codebase, Usage, Context, Deployment, Bugs, History, Roadmap, Ideas) 2. Matching O-entry for session logging (O00XX ↔ P00XX) Example: create_project({ name: 'Carlo Auftrag', tech: 'Python/SAP', description: 'SAP Freigabe-Automatisierung' })

• delete

  Permanently delete a memory entry or sub-node by ID. For root entries: deletes the entire entry with all sub-nodes and tags. For sub-nodes: deletes only that node and its subtree. ⚠️ IRREVERSIBLE — no undo. Use with caution.

• export_memory

  Export your memory, excluding secret entries and secret sub-nodes. Use for sharing, backup, or publishing a sanitized version of your memory.

• find_related

  Find entries related to the given entry. Uses tag overlap first (intentional connections, marked [T]), then FTS5 keyword matching as supplement (marked [~]). Use to discover connections or spot potential duplicates.

• flush_context

  Store a conversation chunk as linear context history (O-prefix). The AI does the summarization: chunk raw text by topic, then summarize progressively. Recommended: provide L1 (title) + L2 (paragraph summary) + L5 (raw text). L3/L4 are optional intermediate levels for extra detail. O-entries are hidden from bulk reads but discoverable via search, tags, and context_for. Use during /save to preserve raw session context alongside curated P/L/D/E entries.

• import_memory

  Import entries from a .hmem file into your memory. Deduplicates by L1 content (merges sub-nodes), remaps IDs on conflict.

• list_projects

  List all projects (P-entries) with their IDs and titles. Minimal output for checkpoint agents.

• load_project

  Load a project and activate it. Returns L2 content + L3 titles — the perfect project briefing. Also marks the project as active (deactivates any previously active project in the same prefix). Use this when starting work on a project. It combines read_memory(id, depth=3) + update_memory(active=true) in one call. Example: load_project({ id: 'P0048' }) Returns: Overview, Codebase, Usage, Context, etc. with L3 subcategory titles.

• memory_health

  Audit report for your memory: broken links (links pointing to deleted entries), orphaned entries (no sub-nodes), stale favorites/pinned (not accessed in 60 days), broken obsolete chains ([✓ID] pointing to non-existent entries), and tag orphans (tags with no matching entry). Run before/after a curation session.

• memory_stats

  Shows budget status of your memory: total entries by prefix, nodes, favorites, pinned, most-accessed, oldest entry, stale count (not accessed in 30 days), unique hashtags, and avg nodes per entry.

• move_memory

  Move a sub-node (and its entire subtree) to a different parent, updating all ID references. source_id must be a sub-node (e.g. 'P0029.15'), not a root entry. target_parent_id is the new parent: a root entry (e.g. 'L0074') or a sub-node (e.g. 'P0029.20'). Use during curation to reorganize entries into the correct hierarchy.

• move_nodes

  Move session (L2), batch (L3), or exchange (L4) nodes between O-entries. Handles ID rewriting, tag migration, and cleanup of empty parents.

• read_memory

  Read from your hierarchical long-term memory (.hmem). At startup, you received all Level 1 entries (rough summaries). Use this tool to drill deeper into specific memories. Query modes: - By ID: read_memory({ id: 'P0001' }) → L1 + direct L2 children (one level at a time) - By node ID: read_memory({ id: 'P0001.2' }) → that node's content + its direct children - By prefix: read_memory({ prefix: 'L' }) → All Lessons Learned (Level 1) - By time: read_memory({ after: '2026-02-15', before: '2026-02-17' }) - Search: read_memory({ search: 'SSE' }) → Full-text search across all levels - Time-around: re…

• read_project

  Read a project's context without activating it — for cross-project reference. Returns a focused briefing (Overview, Codebase titles, Usage, Context, Requirements titles, Roadmap titles) without session history, rules injection, or changing the active project. Example: read_project({ id: 'P0048' }) Use this when working on Project A but needing context about Project B.

Signals

Stars

24

Forks

5

Open issues

0

Last commit

Jul 21, 2026 (3 months ago)

Last release

May 24, 2026 (5 months ago)

From GitHub (fetched Oct 24, 2026).

Details

Registry name

io.github.Bumblebiber/hmem-mcp

Version

1.6.7

Auth

No auth

Packages

npm hmem-mcp

License

MIT

Repository

github.com/Bumblebiber/hmem

Website

github.com/Bumblebiber/hmem#readme

First seen

Oct 24, 2026

Last checked

Oct 25, 2026

Source

Official MCP Registrygithub-graphql