DEVELOPMENT · AGENT SKILLS DIRECTORY

Mnemoverse Memory MCP

Mnemoverse Memory MCP connects Claude Code, Cursor, VS Code, and ChatGPT to Mnemoverse's hosted persistent memory engine over the Model Context Protocol. The open-source MIT package runs a local stdio adapter or connects via remote OAuth, while memory storage and retrieval run on Mnemoverse's hosted service.

Explore the workflow and limits ↓

What to use it for

Use Mnemoverse Memory MCP when you want decisions, preferences, and project lessons to follow you across multiple AI clients instead of staying locked inside one local CLAUDE.md or .cursorrules file.

At commit a465206 (package version 0.18.1), the server exposes 11 MCP tools: memory_write, memory_read, memory_list_recent, memory_feedback, memory_stats, memory_graph, memory_create_room, memory_invite_to_room, memory_join_room, memory_list_rooms, and vault_list. Rating recalled items with memory_feedback adjusts future recall ranking.

Good fit: developers or teams that want managed cross-client memory with OAuth or API-key access and optional shared rooms. Choose another tool if you require offline or local-first storage where memories never leave your machine.

License, pricing, and hosted boundaries

The TypeScript MCP server (@mnemoverse/mcp-memory-server) and Python SDK are MIT-licensed, while the underlying memory engine is a hosted service:

  • Free tier (no credit card documented): 1,000 successful queries per day (reads, writes, and feedback combined), 10,000 atoms of storage, and 60 requests per minute on mnemoverse.com/pricing.
  • Paid plans: Pro ($29/month), Team ($149/month), and custom Enterprise agreements for self-hosted engine deployments.
  • Not local-first: even when running the MCP package locally via npx, memory payloads and queries are sent over HTTPS to https://core.mnemoverse.com/api/v1.

Setup options (Remote OAuth or Local stdio)

Mnemoverse documents two primary ways to connect MCP clients (Node.js 18+ required for local stdio):

  • Hosted OAuth endpoint: https://mcp.mnemoverse.com/mcp
  • Local stdio with API key (Claude Code example):
claude mcp add mnemoverse --scope user -e MNEMOVERSE_API_KEY=mk_live_YOUR_KEY -e MNEMOVERSE_API_URL=https://core.mnemoverse.com/api/v1 -- npx -y @mnemoverse/mcp-memory-server@latest

Keep API keys in your user-level config rather than a committed repository .cursor/mcp.json. Check the host installation guide for Cursor, VS Code, Zed, and ChatGPT plugin steps.

Scope and operational limits

  • Namespaces vs. access isolation: domain values (such as user:<id>) act as namespaces inside one account, not hard tenant security boundaries.
  • Shared rooms: room members can read room history created before they joined; invitations default to read_write unless explicitly restricted to read.
  • Optional auto-save rules: setup_memory can suggest assistant instructions that save durable facts without prompting each time. Decide what project data is safe to transmit externally before adopting automatic memory-write prompts.
  • Deletion path: the MCP server includes vault_list (returning aliases, purposes, and tags) and memory_feedback, but permanent memory deletion must be performed through Mnemoverse's account/management interface rather than an MCP delete tool.

Primary sources