Claude Code

Claude Code reads .mcp.json at the project root and at the user level. Drop Mneva in either; project-root is per-project (one brain per repo), user-level is global (one brain across everything you do).

Config path

Project: .mcp.json at the repo root
User-wide: ~/.claude/.mcp.json

The format is identical. Use whichever scope you want this brain to apply to.

The paste

{
  "mcpServers": {
    "mneva": {
      "command": "npx",
      "args": ["-y", "mneva-connect"],
      "env": {
        "MNEVA_KEY": "mnv_your_key_here"
      }
    }
  }
}

Replace mnv_your_key_here with the key from mneva.dev/signup.

For multi-agent use (Team plan), add "MNEVA_AGENT": "alice" (or whatever name you want this agent to use on the shared brain) — see Multi-Agent.

Restarting

Claude Code reads .mcp.json at process start. After saving the config:

Project scope — open a new Claude Code session in this repo
User scope — restart Claude Code entirely

You'll see mneva listed in the MCP servers section of /mcp or in the tools panel. The first call invokes npx -y mneva-connect, which pulls the package once and caches it.

Rotating keys

If a key has been exposed (committed by accident, pasted into the wrong tab):

Go to mneva.dev/dashboard
Click Rotate API key
The old key dies immediately; the new key is shown once
Update your .mcp.json
Restart Claude Code

Brain data is untouched. Same memories, same beliefs — different credential.