Email for AI agents. Your AI agents can send email, manage domains, and check delivery status - no SDK needed. One URL, one API key, 20 tools.
Three steps. No code. No SDK. Just a config file edit.
Sign up free and copy your re_live_... key. Free tier works.
Edit your AI client's MCP config. One URL, one Authorization header.
{ "mcpServers": { "sparked-mail": { "url": "https://mail.sparked.network/mcp", "headers": { "Authorization": "Bearer re_live_xxx" } } } }Your agent now knows how to send email and manage domains.
"Send Alice a welcome email from noreply@acme.com"Six example prompts. Each one is a few tool calls the AI figures out on its own.
Agent calls list_templates, picks "Welcome", then calls send_with_template with Alice's name. Done in two tool calls.
Agent calls add_domain, returns TXT records, waits for you to publish DNS, then calls verify_domain.
Agent calls get_message_status, reports delivery state, suggests next steps if it failed.
Agent calls setup_webhook with the URL you give it. Returns the HMAC secret to save on your side.
Agent calls add_suppression with the email. Future sends to that address are blocked at the API level.
Agent calls enable_inbound, returns a one-click Cloudflare Worker deploy URL for you to click.
Pick your AI client, copy the config. That's the whole integration.
// ~/Library/Application Support/Claude/claude_desktop_config.json { "mcpServers": { "sparked-mail": { "url": "https://mail.sparked.network/mcp", "headers": { "Authorization": "Bearer re_live_xxx" } } } }
// Settings -> MCP -> Add new MCP server
Type: http
URL: https://mail.sparked.network/mcp
Headers: Authorization: Bearer re_live_xxx
// ~/.codeium/windsurf/mcp_config.json { "mcpServers": { "sparked-mail": { "serverUrl": "https://mail.sparked.network/mcp", "headers": { "Authorization": "Bearer re_live_xxx" } } } }
Prefer the CLI? Claude Code users can run:
claude mcp add --transport http sparked-mail https://mail.sparked.network/mcp \
--header "Authorization: Bearer re_live_xxx"
Every common email operation is a single tool call. Your agent figures out which to call.
send_emailSend a transactional email
send_with_templateSend using a stored Handlebars template
get_message_statusCheck delivery / open / click state
list_domainsList verified sending domains
add_domainAdd a new sending domain
verify_domainVerify domain via live DNS lookup
list_inboxesList allowed sender local-parts
add_inboxRegister a new sender identity
list_templatesList stored templates
create_templateCreate a Handlebars template
list_suppressionsList blocked recipients
add_suppressionBlock a recipient (unsubscribe)
remove_suppressionUnblock a recipient
get_usageCurrent rate-limit usage
list_recent_eventsRecent delivery events
setup_webhookConfigure webhook URL + secret
create_subaccountCreate a subaccount (agency)
list_api_keysList API keys on the account
enable_inboundEnable inbound (CF Email Routing)
check_spfCheck SPF record status
Your AI agent can't lock you out or destroy data. Destructive operations are REST-only.
Tools like list_domains, get_usage, list_recent_events can be auto-approved by clients - no confirmation needed.
send_email, add_domain, setup_webhook, add_suppression - clients prompt for user confirmation by default.
Delete domain, delete template, rotate API keys, delete account - all REST-only. Agents can't accidentally destroy their own access.