AI tools

Integrate with AI

Use an AI coding assistant to integrate OTP.ps: a ready prompt, the official Agent Skill, the MCP server, llms.txt and OpenAPI.

OTP.ps is built so an AI coding assistant can integrate it for you: read the docs, write a secure server-side integration in your stack, test it in the sandbox, and give you a production checklist. Everything here is free; you only pay for real text messages.

The prompt

Open your project in your AI coding assistant and paste:

Integrate OTP.ps into my application to verify Palestinian mobile numbers using SMS verification codes. Inspect my existing technology stack, follow the official OTP.ps documentation, implement a secure server-side integration, and test it using the sandbox environment before enabling production messaging.

Add your use case at the end, for example "…to confirm cash-on-delivery orders at checkout". Then give the assistant one of the tools below, so it doesn't work from memory.

The Agent Skill

A folder of instructions, references, tested examples and scripts in the open Agent Skills format. It teaches the assistant the whole workflow: inspect the app, choose the architecture, keep the key on the server, normalize numbers, build the send/check endpoints and the form (Arabic RTL and English), handle every error, test with the sandbox numbers, and hand over the checklist. Its scripts check a test key against the sandbox (refusing live keys) and scan the project for leaked keys.

Claude Code, for one project (run in the project's folder):

mkdir -p .claude/skills && curl -fsSL https://otp.ps/skills/otp-integration.tar.gz | tar -xz -C .claude/skills

or for all your projects:

mkdir -p ~/.claude/skills && curl -fsSL https://otp.ps/skills/otp-integration.tar.gz | tar -xz -C ~/.claude/skills

Claude Code then uses it whenever you ask for OTP.ps or phone verification.

Other assistants: if your tool supports Agent Skills, put the same otp-integration folder where it loads skills from (see its documentation). If it doesn't, download the folder into your project and tell the assistant to follow otp-integration/SKILL.md, or tell it to read https://otp.ps/llms.txt.

The MCP server

https://otp.ps/mcp (Streamable HTTP, no account needed). It lets the assistant look things up and try the API in the sandbox while it works:

ToolWhat it does
get_integration_guideA guide for a use case (cash on delivery, registration, passwordless login, phone verification, delivery, appointments) or topic
get_code_exampleTested code for cURL, TypeScript, Node.js, React, Next.js, PHP, Laravel, Python, WooCommerce, Shopify
get_api_referenceThe API reference, or the OpenAPI document
get_error_referenceWhat an error code means and what to do
normalize_phoneValidates and normalizes a number exactly like the API
get_supported_networksNetworks, accepted formats, sandbox numbers
send_test_otp, verify_test_otpSend and check a code in the sandbox: no message is sent, nothing is charged
get_integration_checklistThe production checklist and security practices

It also offers every docs page, the OpenAPI document and the Agent Skill as resources, and an integrate_otp_ps prompt.

Claude Code:

claude mcp add --transport http otp-ps https://otp.ps/mcp

Other clients that support remote MCP servers over HTTP use the same address. For example, a mcp.json-style configuration:

{ "mcpServers": { "otp-ps": { "url": "https://otp.ps/mcp" } } }

and VS Code's .vscode/mcp.json:

{ "servers": { "otp-ps": { "type": "http", "url": "https://otp.ps/mcp" } } }

Without an account, send_test_otp works on the sandbox numbers only, in a private sandbox for your connection. To use any number in your own sandbox, add your test key to the connection, e.g. claude mcp add --transport http otp-ps https://otp.ps/mcp --header "Authorization: Bearer otp_test_…". The MCP server refuses live keys: it can't send real messages or touch your balance.

Your production integration never depends on the MCP server or an AI: it calls the REST API directly.

Account tools (not available yet)

A second address, https://otp.ps/mcp/account, will let an AI tool work on your own account after you approve it:

ToolPermission
send_otp (real messages in live mode; requires an idempotency key)Send codes
verify_otpCheck codes
get_balance, get_usage, get_account_limitsRead balance, usage and limits

The tool signs in with OAuth. You approve it on OTP.ps while signed in to your dashboard, and choose:

  • which permissions it gets;
  • test or live mode;
  • a daily message limit, which is required for live sending.

Everything it does appears in your Activity log, where you can disconnect it at any time.

This is built and tested but switched off until OTP.ps finishes its security review. Until then, that address answers "not found".

Plain documentation for any AI

  • /llms.txt: what OTP.ps is, the key facts, and an index of every page.
  • /llms-full.txt: all the documentation in one file.
  • Any docs page as Markdown: add .md to its address, e.g. https://otp.ps/docs/quickstart.md.
  • /openapi.json: the OpenAPI 3.1 contract, built from the API's code.

What has been tested

MCP serverWith the official MCP client over HTTP (every tool, resource and prompt), and Claude Code connects to it (claude mcp list reports it connected).
Account tools (switched off)The official MCP client completes the whole OAuth sign-in and uses every account tool; Claude Code recognises that /mcp/account needs sign-in.
Agent SkillIts scripts and every example are run or syntax-checked against the API on every change; its references are regenerated from these docs. The folder follows the Agent Skills format checked by the tests.
Other assistantsCursor, VS Code, Codex, Windsurf, Lovable and Replit haven't been tested by OTP.ps yet. The configurations above follow their published formats for remote MCP servers.

View as Markdown