1Password MCP
Secrets for your AI tools, without pasting them into prompts
1Password connects to MCP two ways. The recommended pattern is secret injection: keep API tokens in your vault and resolve them into MCP configs at launch with op:// references or the op run wrapper, so credentials never sit in plaintext config and the AI never sees your vault. There is also a community 1Password MCP server that lets an AI list vaults and look up items via a short-lived Service Account Token — powerful, but remember anything the model reads is sent to your LLM provider, so use a dedicated, read-only vault. Either way it keeps API keys out of prompts, configs and git history.
Install 1Password MCP
Paste into claude_desktop_config.json
{
"mcpServers": {
"1password": {
"command": "npx",
"args": [
"-y",
"onepassword-mcp-server"
]
}
}
}Run in your terminal
claude mcp add 1password -- npx -y onepassword-mcp-server Paste into .cursor/mcp.json
{
"mcpServers": {
"1password": {
"command": "npx",
"args": [
"-y",
"onepassword-mcp-server"
]
}
}
} After adding the server, fully restart your MCP client. Replace <placeholders> (tokens, paths) with your own values.
Facts
- Stars
- ★ 1,800
- Maintainer
- 1Password (official)
Trust score, capability surface (filesystem, shell, network) and risk flags — reviewed in your browser.
Known limitations
Any secret the AI reads is sent to your LLM provider in plaintext and stays in the conversation. For sensitive credentials prefer the injection pattern (op:// / op run) over giving the AI vault-read access. Treat the Service Account Token like a master key and rotate it if exposed.