MCP Server Setup Guide: Let AI Assistants Push Messages Automatically

This page explains how to connect the OnePass MCP Server to popular AI clients (Cursor, Kiro, Claude Desktop, Claude Code, VS Code, JetBrains IDEs, Trae, Gemini CLI, OpenAI Codex CLI, Grok, OpenClaw, Qwen Work, Doubao, Coze, Wukong, DeepSeek Harness and more) so your AI assistant can call the push API and deliver messages to WeChat, DingTalk, Feishu, email and mobile clients automatically.

Why MCP for message push

Model Context Protocol (MCP) is an open standard that lets AI tools call external services. Once a remote MCP server is configured, your AI assistant gains a new capability: sending push notifications. Typical uses include sending task results to your phone after long-running coding jobs, alerting on build or deployment failures, and pushing AI-generated reports to group chats.

Three things every MCP config needs

• Server URL: the streamable HTTP endpoint of the push service.
• Authorization header: most clients require Authorization: Bearer <code> or a custom header.
• Channel code: a 32-character code that identifies your push channel. Replace ****** in every config sample with your own code after signing in.

Each editor tab above provides a ready-to-copy JSON/TOML snippet with the correct root key (mcpServers, servers, context_servers, mcp_servers or mcp.servers) for that client.

Verify the connection

After saving the config, use the client's MCP list command (for example grok mcp list or the servers panel in your IDE) to confirm the server is connected, then ask the assistant to send a test message. If it fails, double-check the channel code and the header name.

What you can do on this page

• Pick your client: the editor tabs switch between Cursor, Kiro, Claude Desktop, Claude Code, VS Code, JetBrains, Trae, Gemini CLI, Codex CLI, Grok, OpenClaw, Qwen Work, Doubao, Coze and the others, each with its own instructions and correct root key.
• Copy the config: every panel has a "Copy config" button on its code block, so you paste the snippet straight into the client's settings file.
• Send the whole skill to an AI: "Copy the full skill to AI" copies the complete SKILL manual, so another agent can install and use the push service for you.
• Use the Agent guide pill: next to the heading, the pill shows the public skill.md address; click it to open, or use its copy / open icons.
• Reuse your channel code automatically: signed-in users get their preset channel code injected into every snippet; a link ending in ?channelCode=... overrides it, which is how you hand a ready-made config to a teammate.
• Test right away: the "Test <channel>" action opens the send-test page with that code, and "Create more channel codes" goes to channel management.
• No-registration channel: without a channel code and unwilling to sign up, the assistant can call create_auth_session to get a third-party login link (Feishu / DingTalk / Gitee / GitHub), or - when you have none of those accounts - an email verification-code page (provider=email, which additionally takes the mailbox and an entry channel code from the public ping list); after you authorize in the browser, poll_auth_session returns your own channel code, and the third-party token never leaves the website. The set of supported login methods is read live from the tool's provider enum and can grow without editing this page.
• Jump to the docs: the link under the heading opens the SKILL / HTTP API documentation.

Note: the page needs login only to auto-fill your channel code - the snippets themselves are public. Your channel code is a credential: pasting it into a config file shared with others means they can push to your channels.