API Docs

FAQs

1. What is MCP (Model Context Protocol)?

The Model Context Protocol is an open standard that lets AI models interact with external tools and services through a common interface. Zwitch's MCP server implements this standard so AI assistants like Claude, Cursor, VS Code, Gemini CLI, and Windsurf can search Zwitch documentation and generate Zwitch integration code directly inside your IDE.

2. What can the Zwitch MCP server do?

It ships 11 tools:

  • 4 documentation tools — search_docs, list_docs, read_doc, zwitch_setup
  • 7 integration-guide generators — payment gateway, account setup, beneficiaries, payouts, collections, verification, webhooks

Languages covered: nodejs, python, go, php. See the Tools Reference for details.

3. Can the AI move my money or call live Zwitch APIs?

No. Every tool is read-only and either searches docs or generates code. The assistant cannot create transfers, run KYC checks, deactivate accounts, or call any Zwitch API on your behalf. The integration code it generates is yours to review, test, and run from your own backend.

4. Do I need a Zwitch account or API keys to use the MCP server?

No. The MCP server itself requires no account and no API keys. You'll need a Zwitch account when you ship the integration code the assistant generates, but not to use the MCP server.

5. Is there a local / Docker version?

No. Zwitch ships only a remote, hosted MCP server. This keeps tool versions and documentation aligned for everyone. If you need a self-hosted deployment for a regulated environment, contact [email protected].

6. Which AI-assisted applications are supported?

Any client that speaks Streamable HTTP MCP. Tested:

  • Claude Code
  • Claude Desktop (via mcp-remote)
  • Cursor (via mcp-remote or native HTTP)
  • VS Code with GitHub Copilot Chat MCP
  • Gemini CLI
  • Windsurf

Each has slightly different config syntax — see Remote MCP Server Setup.

7. Can I use it in production?

Yes — the documentation and code-generation tools are official Zwitch surfaces and safe to use in any environment. Note that the assistant only generates integration code; you still review, test, and ship it yourself.

8. What languages and frameworks do the integration guides support?

Languages: nodejs, python, go, php. Each language guide accepts an optional framework parameter (e.g. express for Node.js, flask or django for Python). If you ask without specifying a framework, you'll get an idiomatic default for that language.

9. Does the MCP server host my webhooks?

No. The MCP server generates the webhook handler code in your stack of choice via get_webhook_guide. You host the resulting handler in your own application.

10. Where do I get help?

Related

Updated about 23 hours ago

Updated about 23 hours ago