otp.com in Claude Code
Add the otp.com skill to your project and Claude Code loads the integration steps whenever the task is OTP verification, then writes the send and verify calls into your code.
mkdir -p .claude/skills/otp && curl -fsSL https://otp.com/ai/files/claude-code/SKILL.md -o .claude/skills/otp/SKILL.md - Loaded the otp skill .claude/skills/otp/SKILL.md
- Read the API key OTP_API_KEY · server env
- Added the send callSMSWhatsAppTelegramEmail
- Added the verify call POST /otp/verify
Set up otp.com in Claude Code
Create an API key in the panel (API Keys). Start with an otp_test_ key. Install the skill with the command below, or download the file and save it at .claude/skills/otp/SKILL.md. Ask for the integration. The agent writes the send and verify calls into your code, with the key read from OTP_API_KEY on the server.
mkdir -p .claude/skills/otp && curl -fsSL https://otp.com/ai/files/claude-code/SKILL.md -o .claude/skills/otp/SKILL.md A project skill. Claude Code reads its description and loads the integration steps when the task is about OTP verification.
What to ask once the file is in place
Add phone verification to sign-up with otp.com: send a code when the user submits their number, then verify it.
Read https://otp.com/ai/prompt.md and integrate otp.com OTP verification into this project.
Frequently Asked Questions
Is my API key safe?
Keep the key in an environment variable or a secret on your server (OTP_API_KEY), never in the rules file or the prompt, and never commit it. Use a test key (otp_test_) while you build.
Does the agent see the code?
No. The API never returns the code. The user receives it on their phone or in their inbox, and your app verifies what they type against the otp_id from the send call.
Can I test without sending real messages?
Yes. A test key (otp_test_) runs in the sandbox: send returns a real otp_id and a pending status, but nothing is delivered and nothing is charged. Verify accepts the fixed code 123456 on the default 6-digit length.
Can Claude Code also run the MCP server?
Yes. The MCP server is a separate thing: it lets an agent send and verify codes for real users at runtime. It is not needed to write the integration. See the MCP server docs.