2-minute setup

Give your AI assistant permanent memory
in two commands

Hooks wire memory into every Claude Code message automatically. MCP brings the same memory to Cursor, Windsurf, Claude Desktop, and more. Pick your install path below.

Get your API key, it's free

Hooks vs MCP: what's the difference?

MCP tools only run when your AI client decides to call them, which isn't every time. Hooks run automatically on every message, no decision required.

Recommended
Automatic on every message
Claude Code hooks are shell scripts wired to lifecycle events. They run before Claude reads your prompt and after it responds. Torqon's hooks extract facts from every message and inject relevant context, with no tool call needed from Claude's side.
Fires on every UserPromptSubmit event
Extracts facts from your message in the background
Injects your stored context before Claude reads your prompt
Loads your full memory at the start of every session
One command to install, lives in ~/.torqon/hooks/
Optional
On demand, when your client decides
MCP gives your AI client a set of tools it can call, but deciding when to call them is the client's job. On a typical coding message, it often skips them. MCP is most useful alongside hooks when you want the assistant to explicitly answer a memory query.
Claude must choose to call a memory tool each time
Works in Claude Desktop and any MCP-compatible client
Best for "what do you remember about X?" queries
Not a substitute for hooks on regular workflow messages
Cursor, Windsurf, VS Code, Zed, and Codex all supported

Install hooks

Automatic memory on every message. The setup you actually want.

Recommended
1
Get your API key
Sign up free. No card needed. Copy your key from Settings → API Keys. It looks like tq-xxxxxxxxxxxx
2
Run the installer
Open any terminal and run:
$npx @torqon/mcp@latest hooks --api-key YOUR_KEY
This writes three hook scripts to ~/.torqon/hooks/ and registers them in ~/.claude/settings.json. Your API key goes into the settings env block. No manual environment variables.
3
Restart Claude Code
Close and reopen Claude Code. On the next session start you'll see "Loading Torqon memory..." as your facts load in.
From here, every message automatically extracts facts in the background and injects your stored memory as context before Claude sees your prompt. Nothing left to configure.

Install MCP (optional)

For explicit memory queries in Claude Desktop, Cursor, Windsurf, and other clients.

Optional
MCP is not a replacement for hooks. Claude decides when to call MCP tools, and on a typical coding message it usually doesn't. Start with hooks for reliable automatic memory. Add MCP if you also want Claude to query its memory explicitly on demand.
1
Run the MCP installer for your client
Replace claude with cursor, windsurf, vscode, zed, or codex for other clients.
$npx @torqon/mcp@latest add claude --api-key YOUR_KEY
2
What it writes
The installer adds a torqon entry to your client's MCP config file:
{ "mcpServers": { "torqon": { "command": "npx", "args": ["-y", "@torqon/mcp@latest"], "env": { "TORQON_API_KEY": "YOUR_KEY" } } } }
Config path: ~/Library/Application Support/Claude/claude_desktop_config.json on macOS, %APPDATA%\Claude\claude_desktop_config.json on Windows.
3
Restart your client
Restart Claude Desktop or your chosen client. A torqon tool will appear in the MCP tools panel. Claude can now explicitly query your memory when it decides the tool is relevant.

What changes after install

Three things that weren't happening before start happening automatically.

01
Your facts survive sessions
Project names, stack decisions, preferences, deadlines. All extracted and stored as you work. Every new session loads them before Claude sees your first message.
02
Stop re-explaining your context
Claude already knows your stack, your project name, your preferences. Torqon told it before you said a word. No more spending the first exchange re-establishing what Claude knew yesterday.
03
Fewer tokens, sharper context
Only the semantically relevant facts get injected, not your full history. Median 68% token reduction across 40 real prompts. Up to 95% on simple lookups.
median 68% reduction  |  up to 95% on simple lookups

Ready to connect?

Free to start. Your first 50 fact stores are on us.

Get API key, it's free Read the docs