A Volanea test key can render and log a message, but it never delivers it. That makes a secure email MCP server setup useful for validating an agent workflow before a real recipient can be contacted—and gives you a practical way to keep test and live operations separate.
A secure setup starts with a project-specific credential, a client configuration you can review, and a human approval step for live sends. Use a test key while configuring the MCP connection, inspect the tools exposed to the agent, and move to a live key only when the intended workflow and approval boundary are clear.
For Volanea, Claude Code can connect over Streamable HTTP with a project API key or through OAuth. The test key is safe for checking whether requests render and log correctly, but it has the same access as a secret key. Treat it as a protected credential even though it cannot deliver mail.
Secure email MCP server setup in Claude Code
Start with a project directory where the Claude Code configuration belongs. We document this API-key setup command for Claude Code:
claude mcp add volanea --transport http https://api.volanea.com/mcp -H "Authorization: Bearer sk_test_..."
Replace sk_test_... with a project test key. The URL and Authorization header are also the two values to use when configuring Cursor, Windsurf, or another compatible MCP client. Keep the key out of chat prompts, screenshots, and source control; a checked-in MCP configuration can turn a local test into a credential exposure.
With Volanea, a test key (sk_test_…) renders and logs messages and returns results without delivering email. It does not enforce domain verification, and it has the same access as a secret key (sk_…). A public key (pk_…) is accepted only by POST /v1/events, so it is not a substitute for the server-side key used by an MCP connection. We document these key types and setup options in our MCP documentation.
If your client supports the MCP authorization flow, you can instead configure https://api.volanea.com/mcp without an Authorization header. The client opens a browser for Volanea sign-in and project and test/live-mode selection. Our documentation describes OAuth discovery, PKCE (S256), refresh tokens, and revocation; an OAuth token represents the same tools, limits, and safety rules as an API key.
Choose the right credential and connection method
API-key and OAuth setups solve different operational problems. An API key is direct and straightforward for a controlled development environment; OAuth avoids pasting a long-lived key into a client configuration and can simplify sign-in for a team. Neither choice should be treated as a guarantee of narrower tool access: with Volanea, the OAuth token represents what a key would, including the same tools and limits.
| Connection choice | What you configure | Best fit | Important boundary |
|---|---|---|---|
| API key | MCP endpoint and bearer key | A developer-controlled project setup | Keep the secret server-side and out of prompts and repositories |
| OAuth | MCP endpoint, then browser sign-in | A client that supports MCP authorization | Sign-in does not reduce the available tools or change their safety rules |
| Public key | Not a general MCP credential | Client-side event submission to POST /v1/events | It cannot replace a secret key for other endpoints |
| Test key | MCP endpoint and sk_test_… bearer key | Setup, rendering, and non-delivering test runs | It still has the same access as a secret key |
Our MCP server is available on every plan and provides 23 tools. The documented examples include project status and API reference tools, template and workflow creation, draft campaigns, domain setup, test sends, and delivery checks. Test sends force test mode, campaigns are created as drafts for a human to launch, workflows start paused, and nothing deletes.
Those controls reduce the chance that a first experiment immediately sends or publishes something. They do not make every tool read-only: review what your client exposes and how its approval interface handles consequential actions before connecting a live project.
Validate tokens and protect the server boundary
A client configuration is only one part of authentication. The MCP authorization specification requires servers to validate access tokens, including that each token was issued for that server as its intended audience. Invalid or expired tokens must receive HTTP 401. A server should accept only tokens valid for its own resources; it must not accept or forward an unrelated token as if it were valid for that server.
That distinction matters when an AI client can connect to several services. A token issued for one service is not a general-purpose pass to another. The MCP authorization rules also require clients to send bearer tokens in the Authorization header rather than in a URL query string. These checks belong at the server boundary, not in a prompt or an agent instruction. The MCP authorization specification sets out those requirements.
For a practical setup, use the provider’s documented endpoint and authentication method, then test that the intended project is selected. If you are building your own MCP server rather than connecting a hosted provider, validate each tool input, enforce access controls, limit tool calls, and sanitize tool outputs. Those are requirements in the MCP tool security guidance, not tasks to delegate to the model.
We do not publish per-tool credential scopes for Volanea’s MCP server. Our documentation says that an API key and OAuth token have the same tools, limits, and safety rules, so use project separation and client-side approval controls rather than assuming a credential can be restricted to one tool.
Make test and live email operationally distinct
Separate test and live credentials in your environments, and make test mode the default while developing an integration. With a Volanea test key, a send request can exercise message rendering and logging without contacting a recipient. That is useful for checking template variables, sender formatting, and whether the agent passes the expected values.
A test result does not validate the business decision to send. Before moving to a live credential, inspect the message body, sender, recipient list, personalization fields, and the tool call that will launch or trigger delivery. Keep the live key in the intended production client configuration only after that review.
If you connect through SMTP rather than MCP, the login is also specific: use volanea as the username and the project’s secret API key as the password. Do not use the test key expecting delivery; sk_test_ messages render and log but never deliver. We send through Amazon SES, but our SMTP credential format is the Volanea login, not a separate set of SES credentials.
A conservative release sequence keeps the irreversible part visible:
- Connect with a test key and confirm the selected project.
- Ask the agent to inspect status or prepare a draft rather than send.
- Render a test message and review sender, content, links, and personalization.
- Confirm the recipient list and count outside the model’s own summary.
- Put the live credential into the approved environment and require a person to authorize a real send.
- Review delivery or event results after launch and rotate credentials if they were exposed.
Keep tool calls reviewable
An AI agent can assemble the right-looking message for the wrong audience. For a campaign or workflow, show the human approver the action, sender, subject, audience definition, and recipient count before launch. A request to prepare a welcome campaign should remain a draft until someone checks that the segment actually contains new customers rather than every contact in the workspace.
Our MCP safeguards make that division practical: campaigns are created as drafts, workflows start paused, and test sends stay in test mode. Use those states to establish a review point in your own process. The model can prepare content and configuration; a person should make the decision to contact real recipients.
We publish grouped examples of MCP tools, but not a complete inventory of all 23 tools. Inspect the tools your client actually makes available instead of assuming a category label describes every action. We also recommend recording who can create drafts, approve sending, and change credentials; separating those responsibilities helps a team understand what happened after a mistaken action.
Account for Amazon SES sending limits
Volanea sends through Amazon SES, so SES account state and regional quotas matter when you plan real sending. Amazon SES sending quotas are region-specific and count recipients, not messages: one email addressed to 10 people counts as 10 against the quota. For a workflow that expands a segment into individual recipients, count the audience rather than the number of API calls or campaign objects.
SES sandbox accounts have a maximum quota of 200 messages per 24-hour period and a maximum sending rate of 1 message per second. In the sandbox, sending is limited to verified email addresses and domains or the SES mailbox simulator. These are SES sandbox rules, not a Volanea plan-volume figure. AWS explains sandbox access and production access requirements, and describes how sending quotas work.
Volanea plans use monthly email credits. One credit sends one email to one recipient, or two if the email has an attachment; credits reset at the start of each billing month and do not roll over. When credits run out, sending pauses until the next month or an upgrade, with no overage billing. Check recipient totals and attachments when estimating campaign use, as both affect the number of credits consumed.
Compare cost with the volume you actually send
The Volanea pricing page lists monthly plan-card prices for annual billing alongside comparison-table labels. Starter’s card lists $4 per month and $48 billed yearly; its comparison-table label is Starter ($5). Growth’s card lists $16 per month and $192 billed yearly; the comparison-table label is Growth ($20). The card figures below are the annual-billing monthly prices used in these examples.
| Monthly use case | Plan included volume | Annual-billing calculation | Operational fit |
|---|---|---|---|
| 7,500 emails per month | Starter: 7,500 emails per month | 4 × 12 = $48/year | A project at its listed monthly allowance |
| 40,000 emails per month | Growth: 50,000 emails per month | 16 × 12 = $192/year | A project with room below Growth’s monthly allowance |
Both plans include unlimited contacts and the full MCP server, as well as transactional email, campaigns, and workflows. The plan choice is about sending volume and domain or team needs—not paying separately to enable an MCP tool. We include 1,000 emails every month on the Free plan, with the MCP server available there too. See the Volanea pricing page for plan details.
We use shared Amazon SES infrastructure and do not offer dedicated IPs. Teams evaluating email for regulated data should also know that a HIPAA business associate agreement is available on request at hello@volanea.com, while we do not hold SOC 2 or ISO 27001 certification. Evaluate those facts against your own policy before putting sensitive information into messages or contact records.
Set boundaries for data and delivery
MCP authentication protects the connection to tools; it does not determine how the recipient’s mail system handles delivery. We provide transactional email, campaigns, and automations through one API, with email sent through Amazon SES. Configure sender-domain authentication and review message content separately from agent access. Our email deliverability and sending guides cover the broader sending side of that work.
We document TLS encryption for traffic to our website, dashboard, and API, and separate live and test keys for each workspace. We do not publish a complete inventory of which message content, contact fields, or other customer data are stored, or a complete data-retention schedule. If your deployment requires a field-level data inventory or specific retention terms, get those details from us before moving production data into the workflow.
For email sent through SES, transport security has more than one boundary: the API connection to the provider and the delivery hop to a recipient’s mail server. AWS describes TLS for its SES API and SMTP connections, while downstream delivery behavior is a separate consideration. For sensitive content, decide whether email is an appropriate channel rather than treating a protected MCP connection as protection for every later hop. AWS’s SES security protocol documentation explains the provider-side transport details.
Go-live checklist for an email agent
Before enabling a live email workflow, walk through these checks with both the developer and the person accountable for sending. They turn general security guidance into concrete configuration decisions:
- Connection: Confirm the MCP endpoint and project selected in the client.
- Credential: Start with a test key, keep it out of prompts and source control, and use a live key only in the approved environment.
- Tool surface: Review the tools the client exposes and what each one can change.
- Approval: Require a person to inspect sender, content, audience, and recipient count before a live send.
- Volume: Count recipients, account for attachments, and consider SES region and sandbox state.
- Recovery: Know who can rotate a key and disable the client connection if it is exposed.
- Ownership: Record who prepares drafts, approves sends, and changes credentials.
A small project sending 7,500 emails monthly can use Starter at its listed annual-billing rate; a production workflow at 40,000 monthly emails fits within Growth’s 50,000-email allowance. Those figures help frame a plan choice, while the recipient-level SES quota and the approval boundary determine whether a particular send is safe to run.
FAQ
Does an MCP test key send a real email?
No. A Volanea sk_test_ key renders and logs messages and returns results without delivery. It is intended for validating the connection and message behavior before a live send. Keep it protected anyway: it has the same access as a secret key, even though test messages never reach recipients.
What should I do if an email MCP key leaks?
Rotate or revoke the exposed key, remove it from the prompt, repository, or log where it appeared, and inspect available activity information for unexpected use. Then update only the approved client configuration. If a live key was exposed, pause sending while you restore a credential boundary and review pending campaigns or workflows.
Can I use Volanea with SMTP instead of MCP?
Yes. The Volanea SMTP login uses volanea as the username and the project’s secret API key as the password. That is useful when an application already speaks SMTP and does not need an AI client to call MCP tools. Use a live key only where delivery is intended; test keys do not deliver.
Does connecting with OAuth give an agent fewer email permissions?
Not with Volanea: our documentation says the OAuth token represents the same tools, limits, and safety rules as an API key. Choose OAuth if its browser sign-in and credential-handling flow fits your client, not as a way to grant narrower tool access. Use project separation and human review to control operational risk.
Start with a test project, keep the live-send decision reviewable, and choose a plan based on recipient volume. Start free with 1,000 emails a month.