By default, Claude (and most AI assistants) forget everything the moment a conversation ends. Start a new chat and you're re-explaining your stack, your preferences, and every decision you already made yesterday. This guide shows the simplest way to give Claude a memory that actually persists between chats — no coding, no account, free.
Each conversation is isolated. Claude has no built-in place to write a note to its future self, so nothing carries over from one chat to the next. To fix that, you give it an external memory it can read and write — a small tool it calls to save a fact and later look it up. The clean way to do this is MCP (Model Context Protocol), the open standard Claude Desktop, Cursor, and other clients use to connect to tools.
remember things and recall them in any future chat.That's the whole setup. No signup, no API keys to manage, no library to install.
Create a space on the home page, copy your URL, and add this to your Claude Desktop MCP config:
{ "mcpServers": { "corvid": { "url": "https://corvid.perch-app.workers.dev/mcp?k=YOUR_TOKEN" } } }
Restart Claude Desktop. In any chat you can now say "remember that I deploy on Cloudflare Workers and prefer TypeScript" and, days later in a brand-new chat, "what do you know about how I deploy?" — Claude calls Corvid, finds the note, and answers with it.
Corvid searches memory by keyword, so a lookup works best when it shares words with what was stored. Anyone who has your space token can read that space, so keep the URL private and don't store passwords or secrets in it. If you need encrypted, team-shared, or semantic-vector memory, a heavier tool may fit better — but for giving your own Claude a memory that just works across chats, this is the fastest path.
The same URL works in Cursor (Settings → MCP) and in any MCP client that speaks streamable HTTP. Building your own agent? Skip MCP and call the plain HTTP API directly. Full setup notes are in the MCP guide.