Bounce House

AI agents are increasingly capable of performing multi-step workflows autonomously: signing up for services, testing integrations, monitoring infrastructure, responding to alerts. But they can only interact with systems that expose tools they can call. The Model Context Protocol (MCP) is the emerging standard for this, supported by Claude Code, Cursor, Claude Desktop, and a growing number of AI development tools.

The Bounce House MCP server gives AI agents direct access to the full platform. An agent can create an ephemeral email receiver, sign up for a service using that address, wait for the verification email, extract the link, follow it in a headless browser, fill out any forms, and clean up, all without human intervention. It can also monitor domain health, generate TOTP codes for two-factor challenges, acknowledge alerts, and query suppression lists. If your team is building AI-powered test automation or agentic workflows that interact with the real world, MCP is how the agent gets access to email, webhooks, and the rest of the external message layer.

Installation

pip install bouncehouse-mcp
# or run directly
uvx bouncehouse-mcp

Configuration

Add to your MCP client config (Claude Desktop, Cursor, Claude Code, etc.):

{
  "mcpServers": {
    "bouncehouse": {
      "command": "uvx",
      "args": ["bouncehouse-mcp"],
      "env": {
        "BOUNCEHOUSE_API_KEY": "<your-api-key>"
      }
    }
  }
}

Set BOUNCEHOUSE_API_URL to override the base URL (defaults to production).

Available tools

Receivers

Tool	Parameters	What it does
`create_receiver`	`label`, `channel`, `ttl_seconds`	Create an email, webhook, SMS, or OAuth receiver
`wait_for_message`	`receiver_id`, `timeout`, `since`	Block until a message arrives (no polling)
`get_messages`	`receiver_id`, `limit`	List messages captured by a receiver
`destroy_receiver`	`receiver_id`	Delete a receiver and stop capturing

TOTP

Tool	Parameters	What it does
`register_totp`	`label`, `secret`, `issuer`, `digits`, `period`	Register a TOTP secret for code generation
`get_totp_code`	`secret_id`	Generate the current valid 2FA code

Domains

Tool	Parameters	What it does
`get_domain_health`	`domain_name`	Get health score, DNS findings, and recommendations
`get_recommendations`	`domain_name`	Get specific DNS fixes for a domain
`apply_dns_recommendation`	`domain_name`, `recommendation_id`	Apply a DNS fix through a connected provider

Alerts

Tool	Parameters	What it does
`get_alerts`	`acknowledged`, `limit`	List active alerts (abuse spikes, delivery failures, etc.)
`acknowledge_alert`	`event_id`, `note`	Mark an alert as handled with a note

Auth providers

Tool	Parameters	What it does
`list_auth_providers`		List configured auth provider integrations
`create_auth_provider`	`provider`, `config`	Connect an auth provider webhook
`get_auth_events`	`config_id`, `limit`	List recent auth events (signups, verifications, resets)

Browser flows

Tool	Parameters	What it does
`follow_link`	`url`, `instructions`	Navigate to a URL and return page state (links, forms, buttons)
`complete_oauth`	`receiver_id`, `instructions`	Get verification link from a receiver's message and follow it
`send_flow_action`	`flow_id`, `action`, `selector`, `value`	Click, fill, submit, screenshot, or wait in an active browser session
`close_flow`	`flow_id`, `mark_complete`	Close a browser session, optionally marking the flow as complete

Platform introspection

Diagnostic tools an agent can call when something looks off, or when planning a workflow that depends on what the account or platform supports.

Tool	Parameters	What it does
`get_status`		Diagnostic snapshot of platform health (db, redis, active incidents). Call when you suspect a platform issue, not as a pre-flight check.
`get_usage`		Account-level volume over the last 30 days plus subscription state. `limits` is null until enforcement ships — handle null as "retry on 429s," not "unlimited."
`get_account_capabilities`		Live capability snapshot — channels enabled, DNS provider attribution, auth providers connected, custom-MX domains. Use to plan workflows around actually-configured features.
`get_changelog`	`limit`	Recent platform changes parsed from CHANGELOG.md. Useful when a previously-working workflow starts failing — check what changed.

Agent: create_receiver(label="signup", channel="email", ttl_seconds=3600)
→ Receiver created. Address: a1b2.signup@shard-bo.ing-bo.ing

Agent: [signs up on target site using the receiver address]

Agent: wait_for_message(receiver_id="rx_123", timeout=30)
→ Message received. Subject: "Verify your email"
  Extracted: verification_links: ["https://example.com/verify?token=abc"]

Agent: complete_oauth(receiver_id="rx_123")
→ Following verification link: https://example.com/verify?token=abc
  Page title: "Confirm your account"
  Buttons: "Confirm" id=confirm-btn

Agent: send_flow_action(flow_id="flow_456", action="click", selector="#confirm-btn")
→ Page title: "Account verified!"

Agent: close_flow(flow_id="flow_456", mark_complete=true)
→ Flow completed. Evidence: screenshot, HAR recording

Agent: destroy_receiver(receiver_id="rx_123")
→ Receiver deleted

The complete_oauth tool is particularly useful: it automatically extracts the verification link from the latest message at a receiver and follows it in a browser, combining three API calls (get message, extract link, start flow) into one.

MCP Server