Skip to main content

​
Overview

Memory is only useful where you work. MemoryLake offers several integration paths, all reading and writing the same memory pool: Open Integrations in the console sidebar to see what’s enabled for your account — each available card generates setup commands pre-filled with your project id and API key.

​
OpenClaw / QClaw Plugin

The richest integration: install once and your agent gains Auto-Recall (retrieves relevant memories and document context before answering), Auto-Capture (durable facts from each conversation are stored and merged in the background), and Auto-Upload (files you share in chat land in your project automatically) — plus explicit memory tools and a CLI.

Plugin Reference

Configuration, tools, CLI, skills, and per-agent scoping.

Coding Agents Scenario

The end-to-end walkthrough with day-to-day examples.

​
REST API

Everything the console does, programmatically: projects, document upload/search, memory add/search/trace/forget, and conflict resolution.
  • Base URL: https://app.memorylake.ai/openapi/memorylake
  • Auth: Authorization: Bearer sk-… (create a key)
  • Live spec: https://app.memorylake.ai/openapi/memorylake/api-docs/open-api
The console’s Integrations → API card also generates an AI-ready prompt describing the endpoints, so you can hand your agent the API surface directly. Full documentation in the API Reference.

​
MCP Server

MemoryLake runs a remote MCP server (Model Context Protocol over HTTP, authorized with standard OAuth2) at:
Connect Claude Code with one command — claude mcp add --transport http --scope user memorylake <url> — or point Codex, Cursor, or Claude Desktop at the same URL. You pick the workspace and boundary on the OAuth consent screen, so no secret ever lands in a config file. See Connect via MCP for per-client setup. Connected clients get search and analysis tools over the memories and documents in scope: OpenAI-protocol clients (like ChatGPT) are automatically served simplified search / fetch variants of the same tools — one endpoint adapts to the client.

​
Which Integration Should I Use?

  • You use an MCP-capable client (Claude Code, Codex, Cursor, Claude Desktop) → MCP. One command, OAuth2, nothing to store.
  • You want automatic memory in a coding agent → OpenClaw/QClaw plugin. Auto-recall, auto-capture, and auto-upload beat manual tool calls.
  • You’re building a product → REST API (generally available) or Memory Router (private preview) for call-level memory.
  • You want MemoryLake’s models too → add the Model Router — same key, OpenAI-compatible endpoint, plus the memorylake-models-openclaw plugin for OpenClaw.

​
Next Steps

OpenClaw Plugin Reference

Install, configure, and master the agent plugin

Connect via MCP

Per-client setup for Claude Code, Codex, Cursor, and Desktop