Integrations
Agent frameworks
send0's tools for the Vercel AI SDK, LangChain.js, LangChain for Python and MCP.
send0 ships the same nine tools for each way you build an agent. Names, arguments and output text are identical everywhere, so a prompt that works with one works with all of them.
Vercel AI SDK
@send0/ai-sdk: tools for generateText, streamText and Agent.
LangChain.js
@send0/langchain: structured tools for createAgent and LangGraph.
LangChain for Python
langchain-send0: a toolkit with sync and async tools.
MCP
@send0/mcp: Claude Code, Claude Desktop, Cursor and any MCP client.
Install
npm i @send0/ai-sdk ai zod # Vercel AI SDK
npm i @send0/langchain @langchain/core zod # LangChain.js
pip install langchain-send0 # LangChain for Python
npx -y @send0/mcp # MCP server over stdioTools
| Tool | What it does | Kind |
|---|---|---|
create_inbox | Create an address like signup-agent@send0.email | Creates |
list_inboxes | List the inboxes the API key can use | Read-only |
wait_for_email | Block until a matching email arrives (or arrived in the last minute); returns the extracted code and link | Read-only |
search_messages | Full-text search, or filter by sender (*@github.com) and direction | Read-only |
get_message | Read one message, new text only unless full_text | Read-only |
list_threads | List conversations, most recent first | Read-only |
get_thread | Read a whole conversation, oldest first | Read-only |
send_email | Send a new email, starting a thread | Sends |
reply | Reply in the same thread, with the right In-Reply-To and References | Sends |
Arguments are listed in the MCP tools reference. Tools that take inbox_id fall back to the default inbox you configure.
What every integration does
- Errors become text. An API error comes back to the model as
send0 error <code>: <message>, for examplesend0 error recipient_not_allowed: …, so it can pick another recipient or tell you, instead of the run failing. - Email is untrusted. Tools that return mail wrap each body in
<untrusted_email>tags with a note to treat it as data, and messages flagged for prompt injection carry a warning. Tags inside an email are neutralized, so a sender can't close the block early. - Sending can wait for a person. Each integration can put
send_emailandreplybehind its framework's approval flow, or leave them out entirely. An inbox with theapprovalsend policy turns every send into a draft on the server side too.
Give an agent an API key with read and send scopes, limited to its inbox. See API keys.