Wraith Browser

Advanced Tools

Credential vault, authentication detection, TLS security, AI planning, and autonomous task execution

Advanced tools cover security-sensitive operations and AI-powered planning. The credential vault provides AES-256-GCM encrypted secret storage. Authentication detection identifies login flows automatically. TLS tools manage fingerprint profiles. Planning tools use Monte Carlo Tree Search for action selection, and browse_task runs fully autonomous multi-step browsing.

Credential Vault

The vault uses AES-256-GCM encryption backed by an SQLite database at ~/.wraith/vault.db. All secrets are encrypted at rest. The vault auto-unlocks with an empty passphrase in MCP mode.

ToolDescriptionKey Parameters
browse_vault_storeStore a credential in the encrypted vaultdomain (required), kind (required), identity (required), secret (required)
browse_vault_getRetrieve a credential (decrypts the secret)domain (required), kind
browse_vault_listList all credentials (metadata only)(none)
browse_vault_deleteDelete a credential by IDid (required)
browse_vault_totpGenerate a TOTP 2FA codedomain (required)
browse_vault_rotateRotate a credential's secret valueid (required), new_secret (required)
browse_vault_auditView vault audit log entrieslimit
vault_lockLock the vault and zeroize the master key(none)
vault_unlockUnlock the vault with a passphrasepassphrase
vault_approve_domainApprove a domain to use a credentialcredential_id (required), domain (required)
vault_revoke_domainRevoke a domain's access to a credentialcredential_id (required), domain (required)
vault_check_approvalCheck if a domain is approved for a credentialcredential_id (required), domain (required)

Credential kinds: "password", "api_key", "oauth_token", "totp_seed", "session_cookie", "generic".

Authentication Detection

ToolDescriptionKey Parameters
auth_detectDetect auth flows on a page (password, OAuth, 2FA, CAPTCHA)url

TLS Security

ToolDescriptionKey Parameters
tls_profilesList available TLS fingerprint profiles(none)
tls_verifyVerify TLS fingerprint matches Chrome 136(none)

Planning & Prediction

ToolDescriptionKey Parameters
mcts_planUse MCTS to plan the best next actionstate (required), actions (required), simulations
mcts_statsShow MCTS planner statistics(none)
prefetch_predictPredict which URLs to prefetchtask_description (required)

Autonomous Task Execution

ToolDescriptionKey Parameters
browse_taskRun an autonomous multi-step browsing taskdescription (required), url, max_steps

Examples

Store and retrieve credentials

{
  "tool": "browse_vault_store",
  "arguments": {
    "domain": "github.com",
    "kind": "password",
    "identity": "user@example.com",
    "secret": "my-password"
  }
}
{
  "tool": "browse_vault_get",
  "arguments": {
    "domain": "github.com"
  }
}

Domain approval controls which sites can access a credential:

{
  "tool": "vault_approve_domain",
  "arguments": {
    "credential_id": "a1b2c3d4-...",
    "domain": "login.github.com"
  }
}

AI-powered action planning

{
  "tool": "mcts_plan",
  "arguments": {
    "state": "On job listing page with 10 results visible",
    "actions": ["click @e1", "click @e5", "scroll down", "navigate /page/2"],
    "simulations": 200
  }
}

Returns ranked action recommendations with UCB1 confidence scores. Use this when the agent needs to decide between multiple possible next steps.

Run an autonomous task

{
  "tool": "browse_task",
  "arguments": {
    "description": "Find the top 5 remote Rust jobs on HackerNews Who's Hiring and extract company names and URLs",
    "url": "https://news.ycombinator.com",
    "max_steps": 30
  }
}

Executes a fully autonomous multi-step browsing session using the AI agent loop. The agent decides which pages to visit, what to click, and when the task is complete.

On this page