Shell
mails CLI
Claude Code runs visible commands that are easy to reproduce in a terminal or project runbook.
Direct answer
Can Claude Code read email?
Yes, through a mailbox tool or CLI. Use a dedicated project inbox and keep credentials outside the prompt.
Direct answer
Claude Code can read email by calling the mails CLI through shell tools or by connecting to mails-agent-mcp in an MCP-capable environment. The safer pattern is a dedicated mailbox with a scoped key stored locally. Claude Code can then list inbound messages, search, wait for codes, retrieve messages, and inspect threads without access to a personal human inbox.
Decision map
Use CLI for explicit commands or MCP for tool selection
- 01mails CLI
Claude Code runs visible commands that are easy to reproduce in a terminal or project runbook.
- 02mails-agent-mcp
The client exposes named tools and keeps mailbox state in the local server process.
Two paths
Use CLI for explicit commands or MCP for tool selection
MCP
mails-agent-mcp
The client exposes named tools and keeps mailbox state in the local server process.
Setup
Start without connecting personal email
-
01
Install or configure the local tool
Use the CLI package or add the MCP server command to the client.
-
02
Bootstrap a temporary inbox
Acquire a receive-only address and inspect the mailbox scope and expiry.
-
03
Trigger a known message
Use the address in the repository task or external signup flow.
-
04
Retrieve the bounded result
Ask for the current code, message, or thread rather than all mail.
CLI setup
npm install -g mails-agent
mails bootstrap
mails inbox --direction inbound --limit 5Security
Keep repository, model, and mailbox secrets separate
Do not add the mailbox key to CLAUDE.md, AGENTS.md, environment examples, commits, or chat messages. Store it in local client state or the environment secret mechanism used by the workspace.
A dedicated inbox reduces exposure but does not make email content trusted. Validate links and require approval before sending or changing external systems.
- Use one mailbox per project trust boundary.
- Do not read a human mailbox unless that is explicitly the task.
- Redact message bodies and codes from broad logs.
- Revoke local state when the project ends.
Interface choice
Use deterministic code when Claude does not need to choose
| Task | Recommended interface | Reason |
|---|---|---|
| Interactive repository setup | CLI or MCP | Agent chooses the next email operation. |
| Committed automated test | HTTP or SDK | Control flow and assertions should be deterministic. |
| One-time signup verification | Temporary CLI inbox | Smallest useful scope and automatic expiry. |
| Recurring project notifications | Persistent project mailbox | Stable address and controlled history. |
Implementation judgment
Give the coding agent a tool boundary, not a human account
A dedicated mailbox limits history and revocation to the project task. Keep the key in the local tool process, define the allowed operations, and review any step that turns received content into an external or irreversible action.
Evidence
Sources and product scope
Product behavior is verified against the mails0 source and documentation. External comparisons link to official vendor documentation checked on 2026-08-15.
- mails0 source repositoryImplementation, license, and deployment source
- Claude Code email guideCLI and MCP examples
Questions and answers
Related questions
Does Claude Code need Gmail OAuth to read mails0?
No. mails0 uses a dedicated mailbox-scoped key through its CLI, SDK, API, or MCP server.
Can Claude Code extract an OTP code?
Yes. It can call mails code or the MCP wait_for_code tool for its scoped mailbox.
Can Claude Code send email?
Eligible permanent or self-hosted mailboxes can send through approved tools. Hosted provisional mailboxes are receive-only.
Start with a bounded inbox
Give the agent an inbox, then prove the workflow.
Start with a receive-only temporary mailbox. Move to a permanent identity or self-hosted deployment only after the workflow reaches a real email value event.