How to Secure Environment Variables for LLMs, MCPs, and AI Tools Using 1Password or Doppler

If you're using AI coding tools like Claude Desktop, Cursor, or Continue, you've probably configured MCP servers or API keys somewhere. And if you're like most developers, those secrets are sitting in plain text in JSON config files.

That's a security nightmare waiting to happen—especially with the rise of prompt injection attacks and untrusted MCP servers potentially accessing your environment.

After testing various approaches, I've found two rock-solid solutions: 1Password CLI for those already in the 1Password ecosystem, and Doppler for teams needing centralized secret management. Here's exactly how to implement each one.

Part 1: Understanding the MCP Security Model

When an AI assistant uses an MCP tool, your secrets are injected by the host app as process environment variables only when the MCP server starts. The LLM never sees the values; the MCP server process does. This just-in-time injection is the core security boundary we’ll leverage.

Part 2: The 1Password CLI Approach (Personal Use)

If you're already using 1Password, their CLI provides excellent integration for personal secret management.

# macOS
brew install --cask 1password-cli

# Windows (winget)
winget install 1Password/CLI

# Linux (apt)
curl -sS https://downloads.1password.com/linux/keys/1password.asc | \
  sudo gpg --dearmor --output /usr/share/keyrings/1password-archive-keyring.gpg
echo "deb [arch=$(dpkg --print-architecture) signed-by=/usr/share/keyrings/1password-archive-keyring.gpg] https://downloads.1password.com/linux/debian/$(dpkg --print-architecture) stable main" | \
  sudo tee /etc/apt/sources.list.d/1password.list
sudo apt update && sudo apt install 1password-cli

# Sign in (will prompt for account details)
op signin

# Create a vault for AI tools
op vault create "AI Tools"

# Store secrets
op item create \
  --category "API Credential" \
  --title "Anthropic API Key" \
  --vault "AI Tools" \
  'api_key[password]=sk-ant-...'

op item create \
  --category "API Credential" \
  --title "OpenAI API Key" \
  --vault "AI Tools" \
  'api_key[password]=sk-...'

1Password env template (.env.op)

ANTHROPIC_API_KEY="op://AI Tools/Anthropic API Key/api_key"
OPENAI_API_KEY="op://AI Tools/OpenAI API Key/api_key"
SUPABASE_ACCESS_TOKEN="op://AI Tools/Supabase Token/api_key"
GITHUB_TOKEN="op://AI Tools/GitHub Token/token"

Wrapper and IDE config

1Password needs a wrapper to refresh auth periodically. Create ~/.config/1password/op-mcp-wrapper:

#!/bin/bash
# ~/.config/1password/op-mcp-wrapper
if ! op account get &>/dev/null 2>&1; then
    eval $(op signin)
fi
op run --env-file="$HOME/.env.op" -- "$@"

Make it executable:

chmod +x ~/.config/1password/op-mcp-wrapper

Claude Desktop config (~/Library/Application Support/Claude/claude_desktop_config.json):

{
  "mcpServers": {
    "github": {
      "command": "/Users/YOUR_USERNAME/.config/1password/op-mcp-wrapper",
      "args": ["npx", "-y", "@modelcontextprotocol/server-github"]
    },
    "filesystem": {
      "command": "/Users/YOUR_USERNAME/.config/1password/op-mcp-wrapper",
      "args": ["npx", "-y", "@modelcontextprotocol/server-filesystem", "/Users/YOUR_USERNAME/projects"]
    }
  }
}

✕−+

1Password Access Requested system prompt with Touch ID

Part 3: The Doppler Approach (Team-Friendly)

Doppler is purpose-built for developers environment variables across teams and environments. It's my top recommendation if you need centralized control and audit logging, and is equally great for CI/CD and general development environment focused secrets/automatic injection. They have a lot of deployment options, including normal shell and API access and a CLI tool like 1Password's, but also custom packages for many different languages/frameworks.

# macOS/Linux
curl -Ls https://cli.doppler.com/install.sh | sh

# macOS (Homebrew)
brew install dopplerhq/cli/doppler

# Windows (Scoop)
scoop install doppler

# Login to Doppler
doppler login

# Create a project for your AI tools
doppler projects create ai-tools

# Set up a config for local development
doppler setup --project ai-tools --config dev

# Add your secrets
doppler secrets set ANTHROPIC_API_KEY "sk-ant-..."
doppler secrets set OPENAI_API_KEY "sk-..."
doppler secrets set SUPABASE_ACCESS_TOKEN "sbp_..."
doppler secrets set GITHUB_TOKEN "ghp-..."

# Create a service token for your local machine
# (copy the token printed, it starts with dp.st.)
doppler configs tokens create --project ai-tools --config dev --name "desktop-mcp"

# Store the token in Doppler CLI config (no env var export)
doppler configure set token dp.st.dev.xxxxx

Doppler MCP configuration

Here's how to configure popular MCP servers to use Doppler for secret injection:

Edit ~/Library/Application Support/Claude/claude_desktop_config.json:

{
  "mcpServers": {
    "github": {
      "command": "doppler",
      "args": [
        "run",
        "--project", "ai-tools",
        "--config", "dev",
        "--",
        "npx",
        "-y",
        "@modelcontextprotocol/server-github"
      ]
    },
    "supabase": {
      "command": "doppler",
      "args": [
        "run",
        "--project", "ai-tools",
        "--config", "dev",
        "--",
        "npx",
        "-y",
        "@supabase/mcp-server-supabase",
        "serve",
        "--project-id", "your-project-id"
      ]
    },
    "postgres": {
      "command": "doppler",
      "args": [
        "run",
        "--",
        "npx",
        "-y",
        "@cloudflare/mcp-server-postgres"
      ]
    }
  }
}

Part 4: Comparison & Recommendations

Here's how Doppler and 1Password stack up for AI tool secret management:

When to Use 1Password

• Personal projects where you're the only developer
• Already using 1Password for other passwords
• Highest security requirements (biometric auth on every access)
• Mixed secret types (not just env vars but also passwords, SSH keys)

When to Use Doppler

• Team environments where multiple developers need the same secrets
• CI/CD pipelines that need programmatic access
• Rapid iteration where you're frequently updating secrets
• Multiple environments (dev, staging, prod) with different values
• Audit requirements for compliance

Part 5: Security Best Practices

doppler settings webhook create \
  --project ai-tools \
  --url "https://your-monitoring.com/doppler-webhook" \
  --event "secret.fetch"

# Check recent secret access
op events-api list --limit 10 | jq '.[] | select(.action == "reveal")'

# Revoke all service tokens
doppler configs tokens revoke --all --project ai-tools --config dev

# Generate new token
doppler configs tokens create --project ai-tools --config dev --name "new-desktop"

# Sign out all sessions
op signout --all

# Rotate the specific secret
op item edit "Anthropic API Key" api_key="sk-ant-new-key..."

Part 6: Troubleshooting Common Issues

TL;DR

Securing your environment variables/secrets/credentials doesn't have to be complicated. With 1Password CLI or Doppler, you can:

• Keep all secrets encrypted and centralized
• Inject them just-in-time without exposing them in config files
• Maintain a full audit trail of what accessed which secret when
• Rotate credentials easily when needed

For most individual developers, 1Password is the most convenient choice. For teams and CI/CD, I prefer Doppler.