Clude Memory MCP

MCP server exposing a 4-tier cognitive memory architecture inspired by Stanford's Generative Agents (Park et al. 2023).

Tools

recall_memories

Search the memory system. Returns scored memories ranked by relevance, importance, recency, and vector similarity.

• query — text to search against memory summaries
• tags — filter by tags
• related_user — filter by user/agent ID
• memory_types — filter by type: episodic, semantic, procedural, self_model
• limit — max results (1-20, default 5)
• min_importance — minimum importance threshold (0-1)

store_memory

Store a new memory. Memories persist across conversations, decay over time if not accessed, and get committed to Solana.

• type — episodic (events), semantic (knowledge), procedural (behaviors), self_model (identity)
• content — full memory content
• summary — short summary for recall matching
• tags — tags for filtering
• importance — importance score 0-1
• source — origin identifier (e.g. mcp:my-agent)

get_memory_stats

Get statistics: counts by type, average importance/decay, dream session history, top tags.

get_market_mood

Get current market mood and price state (no LLM call).

ask_clude

Ask Clude a question and get an in-character response. Calls Claude API.

Setup

npm install clude-bot

Requires a Supabase project with the schema from supabase-schema.sql. Set SUPABASE_URL and SUPABASE_SERVICE_KEY environment variables.

Architecture

• 4-tier memory: episodic (7%/day decay), semantic (2%/day), procedural (3%/day), self_model (1%/day)
• Hybrid retrieval: pgvector cosine similarity + keyword matching + tag scoring
• Dream cycles: consolidation, reflection, emergence — every 6 hours
• On-chain commitment: SHA-256 hashed memories committed to Solana via memo transactions
• Granular decomposition: per-fragment embeddings for precise sub-memory retrieval

License

MIT