The email API your agent can run.
Point your agent at mcp.samva.dev or the samva CLI. It sends email, reads the replies, verifies your domain, and edits templates, and you publish the version you reviewed.
The email API your agent wires up for you.
Hand your agent a short brief. It reads the credential guide, connects over MCP or the samva CLI, sends an email, and reads the reply that comes back.
# agent transcript
→ read samva.dev/auth.md
→ connect mcp.samva.dev
← OAuth approved by a person→ tool messages_send_email
{ to: "ada@example.com",
subject: "Your code is 391042" }
← msg_7q2xk9mvt4znw8rh · pending
→ tool conversations_get
← reply from ada@example.comEmail APIs were built for a human at a keyboard.
Setup by dashboard
Domain records, senders, and webhooks mean clicking through screens your agent can't see.
Keys pasted into chat
The agent needs a key, so someone copies a secret into a prompt.
Send-only
The agent sends an email and never sees the reply.
Read auth.md. Connect MCP. Send.
Your agent reads auth.md.
Your agent reads /auth.md and the quickstart, and learns how to get a key.
GET https://samva.dev/auth.md
API key → X-API-Key: samva_sk_live_…
servers, CI, unattended agents
OAuth → samva login a human at a terminal
MCP → mcp.samva.devConnect the hosted MCP.
Add mcp.samva.dev to any MCP client that supports OAuth, or run samva login. A person approves the first credential.
{
"mcpServers": {
"samva": {
"type": "http",
"url": "https://mcp.samva.dev"
}
}
}Send and read.
The agent sends email and reads the conversation it starts. Replies reach your webhook as message.received.
→ messages_send_email
← msg_7q2xk9mvt4znw8rh · pending
webhook message.received
{ conversationId: "conv_…" }What your agent can do.
An agent you run in your own MCP client or the CLI, using an API key you create, can:
- Send and schedule email, in a thread or a new one
- Read delivery events and customer replies
- Add and verify domains and senders
- Edit a template in a shared workspace, show you the diff, and publish the version you approved
- Run, pause, and resume campaigns
- Read and reply from its own mailbox, with sends you can hold for approval
- Manage webhooks and check readiness and usage
Without a key, the server still answers questions about Samva and searches the template gallery.
Samva for ChatGPT, Claude, Cursor, and Grok
Open the Samva plugin in ChatGPT, connect one organization, and start with a read-only readiness check.
ClaudeComing soon
ChatGPT
CursorComing soon
GrokComing soon
Your agent gets an address.
An inbox for your agent
Give your agent its own address on samva.email or your domain, like agent@. It reads, searches, and replies with a key that reaches only that mailbox, and you can hold its sends until a person approves them. Read more
Read the mailbox guide- HOSTED MCPmcp.samva.dev, OAuth or an API key
- GRANTS AND POLICIESsend, or hold for approval
- MAILBOXmailbox.message.received
Build with your agent from here.
Questions from developers.
Can my agent create its own key?
A person creates the first credential, through OAuth in an MCP client or samva login. After that, the agent can create and rotate keys through the API, within the access its own credential holds. Every key belongs to one organization.
Which MCP clients work?
Any client that supports OAuth. Point it at mcp.samva.dev. The server also accepts an API key as a bearer header.
What can the MCP server do?
Sending, replies, delivery events, domains and senders, templates, campaigns, scheduled email, webhooks, usage, and mailboxes. The MCP docs list every tool.
Can my agent have its own inbox?
Yes. Create a mailbox on your samva.email subdomain or a domain you verified, and give the agent a mailbox-scoped key. The agent reads and replies through the API, the CLI, or hosted MCP, and you can hold its sends until a person approves them.
How do I limit what an agent can send?
Give the agent a restricted API key that holds only the resources its job needs, such as sending email without managing domains, and revoke it at any time. Every key belongs to one organization, and its sends run under that organization's spend and daily limits.