Quickstart

Three steps and your agent has a hosted brain it can write to and read from across every session and every tool.

Get an API key

Sign up at mneva.dev/signup. You'll see your API key on the next page — it's shown once and never stored in plain text afterwards. Copy it.

A key looks like:

mnv_c8f92f82050d9b141e72fbb2d60b407df76f527b5f9a042a

If you lose it, you can rotate to a fresh one from the dashboard — the old one dies the instant the new one is minted.

Add it to your agent

Drop this into your coding tool's .mcp.json (the canonical MCP server config — Claude Code, Cursor, Windsurf, and every other modern MCP-capable agent reads this file):

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

Restart your coding tool. That's it — your agent now has sixteen new MCP tools.

For per-client paths and screenshots, see Claude Code, Cursor, Windsurf, or the Anthropic API setup if you're not on MCP.

Verify

Ask your agent to remember something specific:

"Remember that we use Postgres 15 for the user database."

It should call the remember tool. Then in a new session — or a different coding tool — ask:

"Where does our user data live?"

It should call recall (semantic, not keyword) and return the Postgres fact, even though no word in your query matches it. That's the smoke test that the brain is wired.

What your agent gets

Sixteen tools across five surfaces:

• Memory — remember, recall, recall_dreams
• Belief — believe, revise, evidence, evidence_for
• Instinct & calibration — flag, assess, predict, resolve, predictions_open, calibration
• State — state
• Team brain — note_about, outside_views

The unusual ones — dreams, calibration in assess, SPRT-weighted beliefs, allostatic state, multi-agent observation — are the ones no other memory product in this category ships. Worth reading the Concepts pages once.