Secure Firecracker microVM sandboxes for AI agents: network policy, PII & injection guardrails.
MCP server for Declaw — secure sandbox execution for AI agents with network policies, PII scanning, prompt injection defense, and audit logging.
Works with Claude Desktop, Claude Code, Cursor, Windsurf, and any MCP-compatible AI tool.
Add to your MCP config:
{
"mcpServers": {
"declaw": {
"command": "npx",
"args": ["-y", "@declaw/mcp-server"],
"env": {
"DECLAW_API_KEY": "your-api-key"
}
}
}
}
claude mcp add declaw -- npx -y @declaw/mcp-server
Set DECLAW_API_KEY in your environment.
| Tool | Description |
|---|---|
create_sandbox | Create a secure sandbox with configurable security policies |
run_command | Execute a shell command inside a sandbox |
read_file | Read a file from a sandbox |
write_file | Write a file to a sandbox |
list_files | List directory contents in a sandbox |
kill_sandbox | Destroy a sandbox |
list_sandboxes | List all active sandboxes |
When creating a sandbox, choose a security preset:
none — No guardrails. Full internet access.standard (default) — PII scanning + audit logging. Full internet access.strict — PII scanning + prompt injection defense + audit logging + network deny-all.You can also pass allowed_domains to restrict outbound traffic to specific domains:
create_sandbox with template="python", security_preset="strict", allowed_domains=["pypi.org", "github.com"]
| Declaw | Other Sandbox Providers | |
|---|---|---|
| Sandbox execution | Yes | Yes |
| Non-bypassable network controls | Yes | ?? |
| PII scanning | Yes | No |
| Injection defense | Yes | No |
| Full audit trail | Yes | Basic |
| Snapshots | Yes | Varies |
| Multiple templates | 8 built-in | Varies |
| Interactive stdio | Yes | Varies |
| Variable | Required | Description |
|---|---|---|
DECLAW_API_KEY | Yes | Your Declaw API key |
DECLAW_DOMAIN | No | Custom API domain (for on-prem deployments) |
For self-hosted Declaw deployments, set the domain:
{
"mcpServers": {
"declaw": {
"command": "npx",
"args": ["-y", "@declaw/mcp-server"],
"env": {
"DECLAW_API_KEY": "your-api-key",
"DECLAW_DOMAIN": "declaw.internal.company.com"
}
}
}
}
Apache-2.0
Source-derived launch command. Check the maintainer’s required arguments and credentials before running:
npx -y @declaw/mcp-serverMerge this template into ~/Library/Application Support/Claude/claude_desktop_config.json. Keep existing servers. Add any arguments, credentials, and permissions required by the maintainer; this template has not been install-tested.
{
"mcpServers": {
"io-github-declaw-ai-mcp-server": {
"command": "npx",
"args": [
"-y",
"@declaw/mcp-server"
]
}
}
}Restart Claude Desktop completely for changes to take effect. Confirm the server appears connected in the client’s tool list, then try a read-only example from its documentation.
Claude Desktop setup reference@declaw/mcp-servernpmio.github.declaw-ai/mcp-server works with any MCP-compatible client. Copy the config snippet from the Configuration section above and add it to the file shown for your client, then restart the application.
~/Library/Application Support/Claude/claude_desktop_config.jsonRestart Claude Desktop completely for changes to take effect.~/.cursor/mcp.jsonRestart Cursor for changes to take effect..vscode/mcp.jsonReload VS Code window for changes to take effect.~/.codeium/windsurf/mcp_config.jsonRestart Windsurf for changes to take effect..mcp.jsonSave at the project root, then start Claude Code in that project and review the MCP server approval prompt. Keep real credentials out of shared files.