@neruva/mcp

Reliability for AI agents — and for multi-agent operations. One install. Your agents stop repeating mistakes, stop spinning in circles, remember across every session, and get more reliable the longer they run. Every run is replayable bit-for-bit for audit. Bring your own LLM. Free tier, no card.

neruva.io · get a key at app.neruva.io

It just works — that's the point

Neruva is push, not pull. You install it once and it runs silently in the background — no tools to learn, no glue code, no prompting. Before each action it surfaces the past mistake (and the fix that worked) so the agent does it right; it breaks loops before they burn your budget; it records everything for replay. The agent never spends a token deciding to "use memory." The automatic operation is the moat — pull-based memory tools can't match it because they require the agent to stop and ask.

What you get

• Stops repeating mistakes. Before a risky step, the past mistake + the working fix are surfaced automatically — and a known-destructive repeat is blocked so the agent self-corrects.
• Stops spinning. Detects an agent looping or stalling on a failing approach and breaks it, instead of letting it burn turns.
• Remembers across sessions and projects. Persistent memory that builds itself from what your agents do. No re-briefing every morning.
• Provable + replayable. Deterministic from a seed — reproduce any run bit-for-bit, days later, for audit and post-incident review. Export the whole memory as one portable .neruva file. Your data is yours.
• Gets more reliable over time. Learns from your agents' own operational history — without retraining or touching model weights.
• Built for multi-agent operations. Coordinates a fleet of agents, keeps shared memory consistent, and catches an agent that's wrong or lying before it poisons the others. In a swarm, one unreliable agent compounds at every handoff — Neruva is the reliability layer underneath.

Install

# One line — wires the lean auto-pilot into your agent (Claude Code, Claude Agent SDK):
npx -p @neruva/mcp neruva-mcp-install

That installs a tiny native hook (bundled for all 6 platforms — Win/Mac/Linux × x64/arm64) that runs in the background. Set your key and you're done:

{
  "mcpServers": {
    "neruva": {
      "command": "npx",
      "args": ["-y", "-p", "@neruva/mcp@latest", "neruva-mcp"],
      "env": { "NERUVA_API_KEY": "nv_..." }
    }
  }
}

Get a free API key at app.neruva.io (no card). Python users: pip install neruva-mcp.

Optional: the MCP tool surface

The product is the silent auto-pilot — you don't need to call tools. But the MCP server also exposes a small, focused set for clients that want direct access: typed Records (ingest/query/timeline/get/update/delete, GDPR forget, compact, export/import), federated recall (agent_recall), the memory knowledge graph (hd_kg_*), snapshot/restore for replay, multi-agent consensus, and op stats. Bring your own LLM — the substrate stays deterministic and $0/call server-side.

Proof (cited honestly on neruva.io/benchmarks)

Config

Set NERUVA_API_KEY. NERUVA_URL defaults to https://api.neruva.io.

The installer writes lean defaults that just work. Useful overrides:

• NERUVA_AUTO_RECORD=namespace[:ttl_days] — auto-record this agent's activity (fire-and-forget; never blocks).
• NERUVA_PRE_ACTION_THRESHOLD — how confident a past mistake must be to surface (default 0.3).
• NERUVA_PRE_ACTION_DEADLINE_MS — hard cap on background enrichment per action (default 1500; fail-open).
• Multi-agent operations: opt in with NERUVA_AUTO_RESOURCE_CLAIM=1 (file-edit coordination across agents). Off by default — solo agents don't pay for it.

License

MIT