Linksee Memory MCP Server¶

Linksee Memory is a local-first cross-agent memory server that gives multiple AI agents shared persistent memory with a 6-layer structure. It implements an Ebbinghaus-style forgetting curve - memories decay over time unless reinforced - and drift detection that catches when your codebase or decisions diverge from what was previously agreed.

Source: awesome-mcp-servers PR #9079 (discovered July 19, 2026) Category: Knowledge & Memory Author: michielinksee Repo: https://github.com/michielinksee/linksee-memory Status: Active

Why This Matters¶

The problem with most memory solutions is they're either too rigid (exact-match retrieval) or too shallow (single-layer storage). Linksee's 6-layer model mirrors how human teams actually remember things: some facts are hot (active sprint decisions), some warm (project conventions), some cold (historical rationale). The forgetting curve ensures stale information doesn't pollute retrieval - a memory that's never accessed naturally fades, just like in a real team.

The drift detection is the killer feature. It monitors whether your codebase still aligns with past architectural decisions recorded in memory. If it detects drift, it surfaces the original decision context so your agent can reconcile.

Installation¶

# Via npm
npm install -g linksee-memory

# Or run directly
npx linksee-memory

Claude Desktop / Hermes Configuration¶

{
  "mcpServers": {
    "linksee-memory": {
      "command": "npx",
      "args": ["linksee-memory"],
      "env": {
        "LINKSEE_STORAGE_PATH": "~/.linksee-memory"
      }
    }
  }
}

Key Tools¶

Tool	Description
`store_memory`	Write a memory with layer assignment (1-6) and decay rate
`retrieve_memory`	Search across all layers with relevance scoring
`reinforce_memory`	Reset decay timer on an existing memory
`detect_drift`	Compare current state against stored architectural decisions
`list_layers`	Browse memory distribution across the 6 layers

6-Layer Structure¶

Layer	Name	Retention	Example
1	Ephemeral	Hours	Current debugging state
2	Session	Days	What I'm working on today
3	Task	Weeks	Feature implementation details
4	Project	Months	Architecture decisions, conventions
5	Domain	Years	Industry knowledge, patterns
6	Persistent	Forever	Core identity, immutable facts

Integration with CorpusIQ¶

Directly relevant to Hermes' multi-session memory problem. Linksee's forgetting curve + drift detection could complement Honcho (session context) and GBrain (knowledge graph) to create a 3-tier memory architecture: Honcho for hot session state → Linksee for decaying cross-session memory → GBrain for permanent knowledge.