FAQ

Security

Does the AI agent ever see my API keys with Straylight-AI?

No. This is the core guarantee of the zero-knowledge architecture.

When an AI agent calls the api_call MCP tool, it provides a service name and request parameters — but no credential. Straylight fetches the credential from the encrypted vault, injects it as an HTTP header at the Go net/http transport layer, and returns only the response data to the agent.

The credential value is never in a Go variable that the MCP layer handles. It is never in the output returned to the agent. The output sanitizer provides an additional verification pass on every response.

To confirm this yourself: after making an API call through Straylight, ask the AI agent to repeat your API key. It cannot, because it was never told it.

Is it safe to give AI agents access to my API keys?

With Straylight-AI, yes — because the agent never actually receives your API keys. It receives the capability your keys provide, not the keys themselves.

Without Straylight, AI agents commonly gain access to credentials through:

• Reading .env files when asked to debug or run code
• Capturing shell history with commands like curl -H "Authorization: Bearer sk-..."
• Log files that contain credential values
• Prompt injection via malicious API responses that cause the agent to echo secrets

Straylight closes all of these vectors through Claude Code hooks (blocks .env reads), transport-layer injection (credential never in MCP layer), and output sanitization (strips credentials from API responses).

What happens to my credentials if someone else accesses the dashboard?

The dashboard is served on localhost:9470 by default and is only accessible from the machine running Docker. It is not accessible over the network unless you explicitly expose it.

Even if someone accesses the dashboard, they cannot view credential values — the dashboard never displays full credential values, only masked versions (last 4 characters). Credentials are only readable by processes that authenticate to the vault using the AppRole credentials, which are internal to the container.

If you need to secure the dashboard further (shared machine, CI server), you can add HTTP basic auth via a reverse proxy (nginx, Caddy) in front of port 9470.

Can credentials be extracted from the Docker container?

The vault data is stored encrypted (AES-256-GCM) in a Docker named volume. The encryption key is derived and stored separately from the data. Accessing the volume files directly gives you encrypted blobs, not credential values.

Within a running container, credentials are briefly in Go process memory during a request. If you have arbitrary code execution inside the container (which would require a serious Docker escape), you could potentially read process memory. This is a limitation of any in-process secrets management.

The practical risk for individual developers is low. For high-security environments, consider running Straylight on a dedicated machine with full-disk encryption.

Does Straylight-AI send my credentials anywhere?

No. Straylight-AI is entirely self-hosted. It has no cloud component, no telemetry, no analytics, and no callback URLs. The Docker image is built from public source code on GitHub. You can audit every line.

When you make an api_call through Straylight, the request goes directly from the Docker container to the external API (e.g., api.github.com). Nothing passes through any Straylight-controlled server.

Can the AI access my database directly?

No. Straylight proxies all database access through the straylight_db_query MCP tool.

When the AI coding assistant needs to query a database, it calls straylight_db_query with a service name and a SQL statement. Straylight provisions a temporary database user via OpenBao’s database secrets engine, runs the query through an internal connection, and returns the results. The AI receives the query results — never the connection string, admin password, or temporary credentials.

Temporary users are created with SELECT permissions only by default and are automatically revoked when their TTL expires (15 minutes). Even if someone obtained the AI’s context window, there would be no usable database credential in it.

Do I need to configure .claudeignore separately?

Not necessarily, but using both provides the strongest protection.

Straylight’s file firewall (straylight_read_file) automatically redacts secret values from files before returning them to the AI. The file structure is preserved — the AI can see field names and configuration layout — but actual secret values are replaced with [STRAYLIGHT:pattern-name] placeholders.

.claudeignore (and equivalent files for other AI coding assistants) prevents the AI from accessing certain files at all. However, these ignore files have been demonstrated to be bypassable in some circumstances.

Using both layers: straylight_read_file redacts on read, and .claudeignore prevents direct access. The straylight_scan tool can generate ignore file content from its scan results, so you don’t need to write the rules manually.

What secrets can the scanner detect?

The scanner applies 14 pattern categories:

• AWS access keys (AKIA...)
• GitHub Personal Access Tokens (ghp_..., github_pat_...)
• Stripe API keys (sk_live_..., sk_test_...)
• OpenAI API keys (sk-proj-..., sk-...)
• Private keys (PEM-encoded)
• .env variable assignments
• Database connection strings (postgres://, mysql://, mongodb+srv://)
• Generic Bearer tokens
• Generic API key patterns
• Slack webhooks
• SendGrid keys (SG....)
• Twilio tokens
• Basic auth in URLs
• Private key files (by filename: .pem, id_rsa, id_ed25519)

You can add custom patterns in config.yaml under the scanner.custom_patterns key.

Are temporary database credentials safe?

Yes, for several compounding reasons:

• Limited permissions: Temporary users are created with SELECT permissions only by default. They cannot modify, insert, or delete data.
• Short TTL: Credentials expire after 15 minutes. There is no way to extend this beyond the configured max_ttl (default: 1 hour).
• Automatic revocation: When the TTL expires, OpenBao automatically runs DROP ROLE or DROP USER. The credential ceases to exist at the database level — it cannot be used even if someone has the value.
• Never in context: The credential values are never returned to the AI coding assistant. They exist only inside the Straylight container for the duration of the query execution.

Even in a worst-case scenario where an AI session was compromised via prompt injection, the attacker would have no usable database credential. The temporary user would already be expired by the time they could act on it.

Compatibility

Can I use Straylight-AI with Cursor?

Yes. Cursor supports the Model Context Protocol. Add Straylight-AI to your Cursor MCP configuration:

{
  "mcpServers": {
    "straylight-ai": {
      "command": "npx",
      "args": ["straylight-ai", "mcp"]
    }
  }
}

Restart Cursor, and the api_call, exec, check, and services tools will be available.

Can I use Straylight-AI with Windsurf?

Yes. Windsurf supports MCP. Configure it the same way as Cursor, using npx straylight-ai mcp as the command.

Can I use Straylight-AI with any MCP client?

Yes, any client that supports MCP with stdio transport. The MCP server speaks the standard MCP protocol. The command to start it is npx straylight-ai mcp.

Does Straylight-AI work on Windows?

Yes, via Docker Desktop for Windows. The npx straylight-ai installer command works in PowerShell and WSL. The MCP server works with both Windows-native Claude Code and WSL-based setups.

Does Straylight-AI work on macOS?

Yes, fully supported on macOS with Docker Desktop for Mac (Intel and Apple Silicon).

Does Straylight-AI work in CI/CD?

Straylight-AI is designed for developer machines, not CI/CD pipelines. For CI/CD, you typically want to inject credentials via the CI platform’s secret management (GitHub Actions Secrets, GitLab CI Variables, etc.) rather than running a persistent local vault.

That said, you can run Straylight-AI as a service container in Docker Compose-based CI setups if you want consistent credential management across dev and CI environments.

Data and Persistence

What happens if the Docker container restarts?

Your credentials are preserved. They are stored in a Docker named volume (straylight-vault-data) that persists independently of the container. When the container starts, it automatically unseals the vault and all your services are available immediately.

You do not need to re-enter credentials after a restart.

What happens if I run docker rm straylight-ai?

The container is removed but your credentials are preserved in the named volume. Running npx straylight-ai again will create a new container connected to the same volume.

To delete credentials permanently, you must also remove the volume:

docker rm straylight-ai
docker volume rm straylight-vault-data

Can I back up my credentials?

The vault data is in the Docker volume straylight-vault-data, typically at /var/lib/docker/volumes/straylight-vault-data/_data. You can back up this directory, but the data is encrypted — you also need to back up the unseal key to restore access.

Alternatively, use the dashboard export feature (Settings > Export) to download an encrypted backup file.

Can I import credentials from another instance?

Yes, using the dashboard export/import feature. Export from the source instance (Settings > Export), then import on the target instance (Settings > Import). Credentials are re-encrypted with the target instance’s vault key during import.

Setup

How do I add a service that’s not in the 16 templates?

Use the Generic REST API template. It supports:

• Any base URL
• Bearer token, Basic auth, API key header, or API key query parameter authentication
• Custom header name for API key authentication

If you need a template for a commonly-used service that isn’t included, open a GitHub issue and it may be added in a future release.

Can I have multiple instances for different projects?

Yes. You can run multiple containers on different ports:

npx straylight-ai                        # First instance (port 9470)
STRAYLIGHT_PORT=9471 npx straylight-ai   # Second instance (port 9471)

Register each with Claude Code under a different name:

claude mcp add work-straylight --transport stdio -- npx straylight-ai mcp
claude mcp add personal-straylight --transport stdio -- npx straylight-ai mcp --port 9471

How do I rotate a credential?

In the dashboard, click the service card, click Edit, and paste the new credential value. Click Verify & Save. The old credential is overwritten in the vault.

Can I use Straylight-AI without Docker?

Not currently. The Docker container bundles OpenBao vault, the Go backend, and the React dashboard. Running these separately is possible but not officially supported. If you have a strong reason to avoid Docker, open a GitHub issue.

Open Source

Is Straylight-AI open source?

Yes. Straylight-AI is open source under the MIT License. The full source code is available at https://github.com/aj-geddes/straylight-ai.

You can read every line of code, build the Docker image yourself, and fork the project.

Who maintains Straylight-AI?

Straylight-AI is built and maintained by High Velocity Solutions LLC. Contributions are welcome via pull request.

What is the project’s license?

MIT License. You can use it commercially, modify it, distribute it, and use it privately without restrictions. The only requirement is retaining the copyright notice.

How do I report a security vulnerability?

Do not open a public GitHub issue for security vulnerabilities. Email security@straylightai.dev with:

• Description of the vulnerability
• Steps to reproduce
• Impact assessment

We will acknowledge within 48 hours and aim to ship a fix within 14 days for critical issues.