MCP · aifa.works/api/mcp
AIfa memory for any AI agent
Connect Claude, Cursor, VS Code or any MCP client to the memory of your AIfa cabinet with one line of config. Your agent recalls what you discussed with AIfa on any of the four sites — and saves new things into the same memory.
- 1. Log in to your cabinet.
- 2. Get your personal key.
- 3. Paste the ready config into your app.
Claude Code
claude mcp add --transport http aifa-memory https://aifa.works/api/mcp --header "Authorization: Bearer <YOUR_KEY>"
Cursor — ~/.cursor/mcp.json
{
"mcpServers": {
"aifa-memory": {
"url": "https://aifa.works/api/mcp",
"headers": {
"Authorization": "Bearer <YOUR_KEY>"
}
}
}
}VS Code — .vscode/mcp.json
{
"servers": {
"aifa-memory": {
"type": "http",
"url": "https://aifa.works/api/mcp",
"headers": {
"Authorization": "Bearer <YOUR_KEY>"
}
}
}
}Claude Desktop — claude_desktop_config.json
{
"mcpServers": {
"aifa-memory": {
"command": "npx",
"args": [
"-y",
"mcp-remote",
"https://aifa.works/api/mcp",
"--header",
"Authorization:${AUTH_HEADER}"
],
"env": {
"AUTH_HEADER": "Bearer <YOUR_KEY>"
}
}
}
}Four tools
memory_search— search your whole memory by meaningmemory_recent— the latest memory records across all channelsmemory_write— save a fact, a decision or a session summary — AIfa will remember itmemory_about— what this memory is and how it is protected
Security
- The agent sees only the key owner's memory — the email comes from the signed key, never from the request.
- Every tool declares all four MCP safety hints: read-only or write, destructive or not, idempotent, open-world.
- Nothing is ever deleted; an identical record is not duplicated.
- Rate limits per key: 60 calls a minute, including 20 writes.
Prefer memory on your own machine?
The same memory engine is open source (AGPL-3.0) and runs locally over stdio: github.com/MaksimGalatin/padam-nocturne