agent-credentials

Universal credential injector for AI agents. Install it. Everything just works.

One install. Zero config. Every CLI tool gets credentials automatically.

The Problem

You have 5 Google accounts, 3 GitHub tokens, and 20 CLI tools — each reading credentials from different env var names. Every new tool needs manual setup. Multi-account is a mess.

The Solution

npm install -g agent-credentials

That's it. agent-credentials:

1. Scans your machine for existing credentials (AWS, GCloud, GitHub, .env files, keychains)
2. Stores them encrypted in a unified store
3. Injects them into every shell — expanding to ALL known env var names each tool expects

Every CLI tool just works. No manual env var setup. No per-tool configuration.

Quick Start

npm install -g agent-credentials

That's it. Install auto-scans your machine, hooks your shell, and stores everything encrypted. Open a new terminal — every CLI tool just works.

Docker

RUN npm install -g agent-credentials
ENTRYPOINT ["sh", "-c", "eval $(agent-credentials inject) && exec \"$@\"", "--"]

Manual inject (no shell hook)

eval "$(agent-credentials inject)"       # bash/zsh
agent-credentials inject --json          # programmatic

What it does

┌──────────────────────────────────────────────────┐
│                agent-credentials                  │
│                                                   │
│  Scan       → discover creds from env, files,     │
│               keychains, .env, AWS/GCloud config   │
│                                                   │
│  Store      → encrypted, multi-account,            │
│               provider-aware                       │
│                                                   │
│  Resolve    → pick default account, check expiry,  │
│               expand aliases                       │
│                                                   │
│  Inject     → eval "$(agent-credentials inject)"   │
│               sets ALL env vars for ALL tools       │
└──────────────────────────────────────────────────┘

Multi-Account

Connect multiple accounts per provider. Default account is used automatically.

# Add accounts
agent-credentials add google --account [email protected] --token ya29...
agent-credentials add google --account [email protected] --token ya29...

# Set default
agent-credentials set-default google [email protected]

# List
agent-credentials accounts google
# → * [email protected] (default)
#   [email protected]

# Injected env vars:
# GOOGLE_ACCESS_TOKEN=<work token>           ← default
# GOOGLE_OAUTH_TOKEN=<work token>            ← alias
# CLOUDSDK_AUTH_ACCESS_TOKEN=<work token>    ← alias
# [email protected],[email protected]

Auto-Scan Sources

agent-credentials automatically discovers credentials from:

| Source | What it finds | | ------------------------ | ------------------------------------------------------ | | Environment variables | Any existing API keys (GOOGLE_API_KEY, GH_TOKEN, etc.) | | ~/.aws/credentials | AWS access keys and profiles | | ~/.config/gcloud/ | Google OAuth tokens | | ~/.config/gh/hosts.yml | GitHub tokens | | ~/.docker/config.json | Docker registry auth | | .env files | Project-level secrets | | macOS Keychain | Stored service passwords | | ~/.kube/config | Kubernetes tokens |

40+ Providers, 200+ Env Vars

Built-in alias registry covers all major tools:

| Provider | Env vars auto-mapped | | --------- | ----------------------------------------------------------------------------------- | | Google | GOOGLE_API_KEY, GOOGLE_ACCESS_TOKEN, GOOGLE_OAUTH_TOKEN, CLOUDSDK_AUTH_ACCESS_TOKEN | | GitHub | GH_TOKEN, GITHUB_TOKEN | | AWS | AWS_ACCESS_KEY_ID + AWS_SECRET_ACCESS_KEY (always paired) | | OpenAI | OPENAI_API_KEY | | Anthropic | ANTHROPIC_API_KEY | | Stripe | STRIPE_API_KEY, STRIPE_SECRET_KEY | | Slack | SLACK_BOT_TOKEN, SLACK_API_TOKEN | | + 33 more | Full registry → |

Same credential → injected under ALL known env var names. Any CLI tool reads its expected var.

CLI Reference

agent-credentials init               # Auto-hook shell + first scan
agent-credentials inject             # Output env var exports (for eval)
agent-credentials inject --json      # Output as JSON
agent-credentials inject --dry-run   # Show what would be injected (masked)
agent-credentials scan               # Re-scan all sources
agent-credentials add <provider>     # Add a credential manually
agent-credentials remove <provider>  # Remove a credential
agent-credentials list               # List all stored credentials
agent-credentials accounts <provider> # List accounts for a provider
agent-credentials set-default <provider> <account>  # Set default account
agent-credentials check              # Health check all credentials
agent-credentials rotate-key         # Re-encrypt with new key
agent-credentials version            # Print version
agent-credentials uninstall          # Remove store, key, shell hooks

For AI Agent Frameworks

agent-credentials is framework-agnostic. Works with any agent that spawns processes:

| Framework | Integration | | ------------- | ---------------------------------------------------------- | | Claude Code | Shell profile — zero code | | OpenClaw | Shell exec env — zero code | | LangChain | subprocess.Popen(env=json.loads(output)) | | CrewAI | Same as LangChain | | AutoGen | Same as LangChain | | Any framework | eval "$(agent-credentials inject)" before tool execution |

Cloud Storage (Supabase)

Set two env vars and agent-credentials auto-detects Supabase as the backend. All operations persist to your Supabase project instead of local files. Zero re-setup on host rebuild.

# Set env vars — agent-credentials auto-detects Supabase
export SUPABASE_URL="https://your-project.supabase.co"
export SUPABASE_SERVICE_KEY="your-service-key"
export AGENT_CREDENTIALS_KEY="your-64-char-hex-key"

# That's it. All operations now use Supabase.

| Feature | Detail | | -------------- | ---------------------------------------------------------------------------- | | Auto-detection | SUPABASE_URL + SUPABASE_SERVICE_KEY set = Supabase; otherwise local file | | Encryption | AES-256-GCM in Node process. Supabase stores ciphertext only. | | Per-workspace | Row-level security isolates each workspace's credentials. | | Migration | Run migrations/001_credential_store.sql in your Supabase project. | | Caching | 5-minute in-memory cache on listAll(), invalidated on writes. |

Programmatic usage

import {
  SupabaseBackend,
  CredentialStore,
  AesGcmEncryption,
} from "agent-credentials";

const store = new CredentialStore({
  backend: new SupabaseBackend({
    supabaseUrl: process.env.SUPABASE_URL!,
    supabaseKey: process.env.SUPABASE_SERVICE_KEY!,
  }),
  encryption: new AesGcmEncryption(process.env.AGENT_CREDENTIALS_KEY!),
});

Security

| Feature | Detail | | --------------- | ------------------------------------------------------------------------- | | Encryption | AES-256-GCM at rest. Auto-generated key on first run. | | Permissions | Store dir: 0700. Key file: 0600. | | No plaintext | Credentials never stored unencrypted. | | Scan only reads | Scanners only read existing files. Never modify. | | No network | Core package makes zero network requests. Scanners read local files only. | | Open source | Full source available. No telemetry. No phone-home. |

Development

git clone https://github.com/maxtongwang/agent-credentials
cd agent-credentials
npm install
npm test          # run all tests
npm run typecheck # TypeScript strict

MIT License