Alaya Mcp

io.github.SecurityRoninPublished by GitHub account SecurityRoninv0.4.7Updated Oct 28, 2026

Local memory engine for AI agents with knowledge graphs, forgetting, and semantic recall

VerifiedSTDIODesktop onlyDatabasesKnowledge & Memory

Overview

AI-generated overview

Local memory engine for AI assistants: stores conversation episodes in SQLite, builds a knowledge graph, and recalls memories semantically.

What you can do
  • Store and search memories: save conversation episodes and run semantic recall across them
  • Manage knowledge: store pre-extracted facts and relationships, browse extracted knowledge, and explore the knowledge graph
  • Inspect memory state: view system health and stats, emergent category taxonomy, and crystallized user preferences
  • Run maintenance: trigger lifecycle operations such as strengthen, transform and forget
  • Configure the embedding provider and other options
When to use it
Useful when you want an assistant to keep durable, structured memory across sessions on your own machine instead of re-injecting full context, and to let weak memories fade while strong ones persist.
Requirements
Runs locally as an npm package (alaya-mcp) over stdio, launched for example with npx. Data is kept in a local SQLite database, by default under the user's home directory, with the path configurable via ALAYA_DB. Optional ALAYA_LLM_API_KEY, ALAYA_LLM_API_URL and ALAYA_LLM_MODEL enable LLM-powered knowledge extraction; without them the assistant extracts knowledge itself and search falls back to BM25.
Before you install
The optional ALAYA_LLM_API_KEY is a secret stored in the client configuration; if set, memory content is sent to the OpenAI-compatible endpoint given in ALAYA_LLM_API_URL, so choose a provider you trust. The server writes and maintains a local database, and lifecycle maintenance can forget memories. Desktop clients only.

Installation

In SourceWeft

  1. Open Alaya Mcp in the dashboard and add it to a workspace.
  2. Enable the server for the chats that should use its tools.

Desktop only via STDIO. STDIO servers start a local process, so they need the SourceWeft desktop host.

Other MCP clients

Follow the launch instructions in the repository.

README

alaya-mcp

The only memory engine with neuroscience-grounded memory dynamics — Bjork dual-strength forgetting, retrieval-induced suppression, and Hebbian co-activation — in a zero-dependency embeddable Rust library.

Stores conversation episodes, consolidates knowledge through a cognitive lifecycle, and builds a personal knowledge graph — all in a local SQLite database.

Install

Claude Code

Add to ~/.claude/claude_code_config.json:

json
{  "mcpServers": {    "alaya": {      "command": "npx",      "args": ["-y", "alaya-mcp"]    }  }}

Claude Desktop

Add to your Claude Desktop MCP config:

json
{  "mcpServers": {    "alaya": {      "command": "npx",      "args": ["-y", "alaya-mcp"],      "env": {        "ALAYA_LLM_API_KEY": "sk-...",        "ALAYA_LLM_API_URL": "https://api.openai.com/v1/chat/completions",        "ALAYA_LLM_MODEL": "gpt-4o-mini"      }    }  }}

The ALAYA_LLM_* env vars are optional — they enable automatic knowledge extraction. Without them, the agent extracts knowledge from its own context.

MCP Tools

ToolDescription
rememberStore a conversation episode
recallSemantic search across memories
statusMemory system health and stats
knowledgeBrowse extracted knowledge
learnStore pre-extracted facts and relationships
categoriesView emergent category taxonomy
preferencesTrack crystallized user preferences
neighborsExplore the knowledge graph
lifecycleTrigger maintenance (strengthen, transform, forget)
configureSet embedding provider and other options

How It Works

  1. Remember — store conversation episodes as they happen
  2. Consolidate — extract facts, relationships, and concepts from episodes
  3. Strengthen — co-retrieved memories strengthen their connections (Hebbian LTP); retrieving memory A actively suppresses competing memories B and C (retrieval-induced forgetting)
  4. Categorize — emergent categories form automatically from your knowledge
  5. Forget — Bjork dual-strength decay separates storage strength (how well-encoded) from retrieval strength (how easily found); weak memories fade, strong ones persist
  6. Crystallize — implicit preferences emerge from accumulated impressions (vasana), no LLM required

All data stays on your machine in ~/.alaya/memory.db.

Why Alaya

ProblemFile-based memoryAlaya
Token wasteFull-context injection (~35K tokens/msg)Ranked retrieval — only top-k relevant memories
No structureEverything in one fileThree typed stores: episodes, knowledge, preferences
No forgettingFiles grow until manually curatedBjork dual-strength decay + retrieval-induced forgetting
No associationsFlat files, no linksHebbian co-retrieval graph (LTP/LTD)
Brittle preferencesAgent-authored summary, driftsImplicit preferences emerge from impressions (vasana)
LLM requiredCan't function without oneGraceful degradation: no embeddings → BM25-only

Links

License

MIT

[Image]

Source: npm/alaya-mcp/README.md at commit 7e43e0b

Tools

0
SourceWeft has not started this package in its sandbox yet, so its tools are not known.

Version history

1
  1. v0.4.7LatestSep 16, 2026

Score

50% of the applicable checks met

How complete this listing is, computed from the registry entry, the endpoint and public sources. Not a rating of the server and not a statement that it is safe or working today: reachability is a dated check, and no installs or reviews go into the score.

  • Reached in the last 30 days25 pts

    Keep the endpoint answering the MCP handshake; SourceWeft probes it every 30 days.

  • Auth declared10 pts

    Declare the credential headers in the registry entry, or keep the endpoint answering the probe so its auth can be classified.

  • Tools listed15 pts

    Answer tools/list without a credential, or ship a package that starts in a sandbox without one.

  • DescriptionRequired15 pts
  • Logo10 pts
  • Documentation15 pts
  • License10 pts