Agentsim MCP server
{
"mcpServers": {
"agentsim": {
"type": "http",
"url": "https://mcp.agentsim.dev/mcp"
}
}
}{
"mcpServers": {
"agentsim": {
"type": "http",
"url": "https://mcp.agentsim.dev/mcp"
}
}
}{
"mcpServers": {
"agentsim": {
"type": "http",
"url": "https://mcp.agentsim.dev/mcp"
}
}
}{
"servers": {
"agentsim": {
"type": "http",
"url": "https://mcp.agentsim.dev/mcp"
}
}
}{
"mcpServers": {
"agentsim": {
"type": "http",
"url": "https://mcp.agentsim.dev/mcp"
}
}
}file .mcp.json (in your project root)file claude_desktop_config.jsonfile ~/.cursor/mcp.jsonfile .vscode/mcp.jsonfile ~/.codeium/windsurf/mcp_config.jsonversion 0.29.1
What it does
MCP server that exposes authentication challenge tools for AI coding assistants. It provides SMS OTP, email OTP, magic links, and WebAuthn test capabilities, allowing developers to programmatically control authentication flows in their AI agent workflows.
- Test AI agents against SMS OTP authentication by provisioning temporary US phone numbers and retrieving OTP codes
- Verify that coding assistants can handle multi-factor authentication challenges during automated testing
- Develop and debug authentication logic in AI applications by controlling when and how authentication verdicts are delivered
Tools this server exposes
| Tool | Does |
|---|---|
| open_challenge | Open an authentication challenge session with a specified channel (sms_otp, email_otp, magic_link, or webauthn_required), returning a session ID and identifier |
| wait_for_verdict | Long-poll for the authentication challenge result, returning the outcome including OTP code, magic link, WebAuthn requirement, or policy denial |
| get_messages | List raw SMS messages received on an active session |
| release_number | Release a session early and return its allocated phone number to the pool |
| list_numbers | List all active challenge sessions for the current account |
| provision_number | Legacy alias for open_challenge with SMS OTP channel |
| wait_for_otp | Legacy alias for wait_for_verdict |
Before you enable it
An MCP server runs with whatever access you give it. Start with a development account rather than production, and check which tools the server exposes before you let an agent call them unattended.