SDKs & Libraries
Official, installable clients for the MailFlat API: one line to install, one line to use. Pick your language or AI framework.
All packages
| Package | For | Install |
|---|---|---|
| mailflat | Python | pip install mailflat |
| @mailflat/sdk | JavaScript / TypeScript | npm i @mailflat/sdk |
| mailflat-mcp | AI assistants (Claude, Cursor) | uvx mailflat-mcp |
| @mailflat/ai-sdk | Vercel AI SDK agents | npm i @mailflat/ai-sdk |
| mailflat.langchain | LangChain agents | pip install "mailflat[langchain]" |
| mailflat-sdks | Java / Selenium | JitPack, see below |
They are all the same API
Every client is a thin, typed wrapper over /api/v1, so behaviour is identical across languages. Authenticate with an account key (mf_live_…) from Agents → API keys.
Install & first call
# install
pip install mailflat
# use: create an inbox and wait for the OTP
from mailflat import MailFlat
mf = MailFlat(api_key="mf_live_…")
inbox = mf.create(label="signup")
otp = inbox.wait_for_otp(timeout=30) # → "482913"
Line by linewhat each step of this Python example does
- mf.create(label=…)
- Opens a plain-text agent inbox and returns its address.
- wait_for_otp(timeout=30)
- Polls /latest for you and raises on timeout instead of returning None.
Which one should I pick?Test suite · agent framework · assistant
• Writing tests: the Python or JS client. Java if your suite is Selenium/JUnit.
• Building an agent in code: the Vercel AI SDK suite or the LangChain toolkit, so the model calls the tools itself.
• Using Claude Desktop or Cursor: the MCP server, no code at all.
• Any other language: the raw API is six endpoints, see Agent API & MCP.
Java source: github.com/onderyentar21/mailflat-sdks (packages/java-sdk).