Evermemos Mcp

What is Evermemos Mcp?

Evermemos Mcp is a long‑term memory server for AI coding assistants. It stores and retrieves context across sessions using the EverMemOS Cloud API, eliminating the need to re‑explain project details every session.

How to use Evermemos Mcp?

Get an API key from EverMemOS Cloud, then add the server to your MCP client configuration with the EVERMEMOS_API_KEY environment variable. Run via uvx evermemos-mcp@latest. Works with Claude Code, Cursor, Cline, Cherry Studio, OpenClaw, Gemini CLI, Aider, and any MCP‑compatible client.

Key features of Evermemos Mcp

• Space isolation with domain:slug per project or topic
• Multi‑space search across up to 10 spaces in one call
• Sensitive content guard blocking API keys, passwords, tokens
• Memory conflict detection in chat spaces
• Lifecycle tracking: queued, provisional, fallback, searchable
• Git auto‑detection infers space from remote repository name

Use cases of Evermemos Mcp

• Store architecture decisions and recall them in a new session
• Save personal preferences like UI theme and keybindings across chats
• Keep cross‑session learning notes under a dedicated study space

FAQ from Evermemos Mcp

How does Evermemos Mcp differ from other memory MCP servers?

It offers space isolation, lifecycle tracking, sensitive content guard, conflict detection, multi‑space search, six retrieval strategies, and benchmark‑verified 60/60 recall with zero attribution errors.

What are the runtime requirements?

Python, uv (or uvx), and a valid EverMemOS Cloud API key set in the EVERMEMOS_API_KEY environment variable.

Where are memories stored?

All memories live on EverMemOS Cloud. There is no local persistence; the server is cloud‑first.

What does the flush parameter do on remember?

flush controls whether the write is immediately sent for extraction. Use false mid‑conversation and true at end of session or topic switch.

Is memory deletion instant?

No, cloud deletion is async and best‑effort. The server provides a verification workflow: confirm the target via fetch_history, call forget, then verify with fetch_history again.