Getting Started

Installation

Install the 0Latency MCP server globally via npm:

Or use it directly with npx (no installation required):

Authentication

Get your API key from 0latency.ai. How you supply it depends on how you are using 0Latency:

CLI, SDKs and direct API calls — export it in your shell:

Claude Desktop — do not rely on this export. Claude Desktop is launched by the operating system, not from your shell, so it never sees variables exported in a terminal. It reads the key from the env block inside claude_desktop_config.json instead — see MCP Setup below.

Your API key is used to authenticate all requests to the 0Latency memory API.

First Memory

Python SDK

Store and recall memories using the Python SDK:

from zerolatency import Memory

# Initialize the client
memory = Memory(api_key="zl_live_...")

# Store a memory
result = memory.seed(
    agent_id="assistant-1",
    facts=[{
        "text": "User prefers concise responses",
        "category": "preference"
    }]
)

# Recall memories
memories = memory.recall(
    agent_id="assistant-1",
    query="How should I format my responses?"
)

print(memories)

JavaScript SDK

Use the JavaScript SDK in Node.js or browser environments:

import { Memory } from '@0latency/sdk';

// Initialize the client
const memory = new Memory({ apiKey: 'zl_live_...' });

// Store a memory
await memory.seed({
    agentId: 'assistant-1',
    facts: [{
        text: 'User prefers dark mode',
        category: 'preference'
    }]
});

// Recall memories
const memories = await memory.recall({
    agentId: 'assistant-1',
    query: 'What are the user settings?'
});

console.log(memories);

MCP Setup

Connect 0Latency to Claude Desktop, Claude Code, Cursor, Windsurf, or any MCP-compatible client.

Recommended — let the package configure it for you

The server ships an init command that does the whole setup, including finding the config file:

export ZERO_LATENCY_API_KEY=zl_live_...
npx -y @0latency/mcp-server init

It asks which client you are configuring, writes the correct JSON to the correct path for that client, and then verifies a real memory write and recall against the API before it finishes — so you know it works before you restart anything. If you have not exported the key it will prompt for it.

Typical run:

? Which client are you configuring? Claude Desktop
✓ Found API key
✓ Config written to: ~/Library/Application Support/Claude/claude_desktop_config.json
⏳ Verifying memory write+recall...
✓ Memory verification: write+recall succeeded in 1.5s
✓ Restart Claude Desktop to apply changes

Alternative — edit the config by hand

Only needed if you would rather not run init, or your client is not in its list. Edit claude_desktop_config.json. It lives here:

If the file does not exist yet, create it. You can also open it straight from the app: Claude Desktop → Settings → Developer → Edit Config.

Add this to it:

{
  "mcpServers": {
    "0latency": {
      "command": "npx",
      "args": ["-y", "@0latency/mcp-server"],
      "env": {
        "ZERO_LATENCY_API_KEY": "zl_live_..."
      }
    }
  }
}

Verify Connection

Restart Claude Desktop and verify the MCP server is connected:

• ✓ Look for the 0Latency icon in the toolbar
• ✓ Try: "Store this memory: I prefer Python over JavaScript"
• ✓ Try: "What programming language do I prefer?"

Next Steps — Choose Your Path

Get started based on your use case:

All Integrations

0Latency works with your existing tools and frameworks: