Disposable email for AI agents: instant inboxes, verification-code and structured-intent extraction (OTP, magic link, billing, security alerts), including codes hidden in .txt and PDF attachments. Task-scoped inboxes with long-poll wait, scoped sub-tokens bound to named agent identities, and one-call self-serve onboarding for new agents.
Scored across 14 tools
Most tools target distinct resources/actions, but await_verification and wait_for_code overlap heavily — both wait for a matching mail and reuse the same wait path, differing mainly in whether they also provision/reclaim the inbox. extract_fields, the extraction built into list_messages/get_message, and manage_extract_rules also sit close together, though the ad-hoc vs. stored-distinction is spelled out in the descriptions.
All names are snake_case verb_noun (list_inboxes, get_message, delete_message, send_email, manage_webhooks), which is predictable and readable. Minor deviations: 'await' vs 'wait' used for near-identical operations, and singular/plural is mixed (create_inboxes vs delete_inbox).
14 tools is squarely in the well-scoped range, and each maps to a real capability of an email/inbox service (provisioning, reading, extraction, webhooks, rules, sending). Nothing looks padded or redundant-by-design beyond the noted wait overlap.
Core inbox and message lifecycle is covered (create/list/delete, list/get/delete messages, usage, send). However, attachment download is only an HTTP endpoint referenced in get_message rather than a tool, there is no inbox pause/resume despite wait_for_code describing paused-inbox behavior, and no restore for the recycle bin — real dead ends for an agent.
One-call verification loop: create an inbox, return its address, wait for a matching verification mail, return the structured result (extracted + actions), and optionally reclaim the inbox. Reuses the same wait path as wait_for_code (15-minute history scan + live wake-up, broadcast semantics). Params: timeoutSec (default 60, max 300), from/subject filters, fromMatch (any|envelope), optional localPart/domain/displayLabel, reclaim (default false — soft-deletes the inbox afterwards; its address enters a 7-day reuse cooldown).
Inputs: from, domain, reclaim, subject, +4 more.
Create one or more temporary inboxes (subject to the plan inbox quota; 10 per call). Returns the created InboxDto objects; per-item failures (e.g. taken addresses) are listed in errors. The token-bound project is used; an optional projectId is accepted but must match it (mismatch is an explicit error). B1 task inboxes: pass taskTtlSec (+ expectFrom and/or expectSubject, optional completeOnMatch) with count=1 to declare an ephemeral task — the inbox auto-reclaims at expiry (expired) and, with completeOnMatch, the first matching mail flips it to completed with matchedMessageId; query via list_inboxes taskStatus filter.
Inputs: count, domain, localPart, projectId, +6 more.
Soft-delete an inbox: it stops receiving mail and its address enters a 7-day reuse cooldown before it can be recreated.
Inputs: inboxId (required).
Soft-delete a message (moves it to the recycle bin).
Inputs: inboxId (required), messageId (required).
Apply ad-hoc named regexes ({fieldName: JS regex source}) to a message body and return the first capture/match per field.
Inputs: fields (required), inboxId (required), messageId (required).
Fetch one message in full, including the structured extracted object and the attachment index (attachments: [{index, filename, contentType, size}], always an array — one unreadable attachment never hides the others). Download files via GET /v1/inboxes/:id/messages/:msgId/attachments/:index. Fetching marks the message as read.
Inputs: inboxId (required), messageId (required).
Current plan limits and usage counters (monthly inbound count, inbox usage).
Inputs: no named parameters.
List the principal inboxes visible to the API token (scoped to the token project). Paginated: limit/offset, response echoes both.
Inputs: limit, offset, projectId, taskStatus.
List messages of an inbox, newest first, with built-in extraction results embedded. Pass nextCursor as before to fetch the next page. With filter.from/subject, the server scans only the newest 100 messages before substring-filtering — matches beyond that require paging with before. An empty result with a non-null nextCursor means more pages exist (do not stop on partial pages).
Inputs: inboxId (required), limit, before, filter.
List, create or delete custom extraction rules for the token project. create requires a paid plan (free tier gets isError, code=quota_exceeded; rule count per plan); list/delete work on every plan. list returns at most the first 100 rules. create: name + fields ({fieldName: JS regex source}), optional scopeInboxId/matchFrom/matchSubject; delete: ruleId.
Inputs: action (required), name, fields, ruleId, +3 more.
List, create or delete webhook endpoints for the token project (endpoint count per plan). list returns at most the first 100 endpoints. create: https url (localhost/127.0.0.1/::1 allowed for testing; loopback/private/reserved hosts and unresolvable domains are rejected — SSRF defense); delete: webhookId. The create response contains the signing secret — shown once.
Inputs: action (required), url, events, webhookId.
Send an outbound email from one of your inboxes. Paid plans only (free tier is rejected with isError, code=quota_exceeded); per-plan daily quota (starter 50 / pro 200 / max 1000). The shared MCP rate limit (120/min) applies; there is no separate send-specific limit on this tool. Recipient must be a valid address; the sending domain must be verified with the email provider.
Inputs: to (required), inboxId (required), subject (required), html, +1 more.
Translate an email into Simplified Chinese or English (Workers AI, on-platform). Best effort: verification codes, URLs and addresses are asked to be preserved verbatim by the prompt. Result is cached per language; long emails may be truncated.
Inputs: target (required), inboxId (required), messageId (required).
Wait for a message with an extracted code/link in an inbox (long-poll, no client polling). Broadcast semantics: multiple concurrent waiters all match the same arriving mail — coordinate multi-agent consumption separately. Mail held while the inbox is paused is matched too. First scans mail that arrived in the last 15 minutes (WAIT_HISTORY_WINDOW) — an already-arrived code returns immediately; otherwise parks until timeoutSec. fromMatch: "any" (default) matches the display From header or the SMTP envelope sender; "envelope" matches ONLY the envelope sender (unforgeable). Note for "envelope": ESP mail (e.g. SES) often uses a bounce address as the envelope sender, so filtering by the brand display domain will not match — use "any" unless you know the envelope domain.
Inputs: inboxId (required), filter, fromMatch, timeoutSec.
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Add one secure layer between your agents and this server.