Workspace Agent 49 MCP tools

Email verification for CI and AI agents

Disposable inbox per CI run — create, wait, and extract OTP or magic link in one call. Scoped run keys, no shared test mailbox, no human OTP checks.

Open console Start QA setup Agent docs

              MCP
              Playwright
              API
              CLI
            

              // Cursor / Codex MCP
await mailagent_verify_signup({
  service: "github",
  runId: "ci-4821",
  deleteAfterSuccess: true,
});

// → agent.primaryAction
// → verification.primaryButton
// → confidence: "high"
            

Platform

Verify signup and workspace mail

Disposable inboxes and structured extraction for tests — plus Workspace Agent for MailAgent inbox messages, reminders, and policy-gated replies. One MCP server for QA and autonomous mail.

Verify signup in one call

Create inbox, wait for the message, extract OTP or magic link, return agent.primaryAction — no inbox UI scraping.

Agent docs →

MCP for Cursor & Codex

MCP tools: verify, workspace mail, simulate, scoped keys.

Browse tools →

Playwright & Cypress

Per-worker labels, subjectContains, simulate without SMTP in CI.

QA guide →

Self-host or hosted

MIT on Cloudflare Workers + Resend, or api.webmailagent.com with an API key.

Integration guide →

Scoped run keys

Agents mint short-lived keys per run — no permanent secrets in CI.

Scoped keys →

Cloudflare Workers
MCP · 49 tools
Resend Inbound
Forwarded mail
DeepSeek · Qwen · Ollama

No SaaS markup
for CI verification

Per-seat QA dashboards charge for UI your CI never opens. MailAgent returns structured OTP and links — not another inbox tab.

Resend Inbound MailAgent

MCP and REST extraction on Workers — self-host with MIT, or api.webmailagent.com when you want hosted speed.

How it works

Up and running in 1, 2, 3…

No shared test inbox. No manual OTP checks.
Scoped run keys instead of permanent secrets in CI.

Create inbox

Test or agent calls MCP/API. You get an isolated address and optional label tied to the CI run.

Wait & extract

Submit the address in signup, then one call returns OTP or magic link as agent.primaryAction — no inbox UI scraping.

Ship in CI

Simulate without SMTP in pipelines. Delete the inbox when the run succeeds, or keep it open for debugging.

Output

Return the action, not raw email markup

Tests and agents get structured verification: OTP, magic link, safe HTML CTA, confidence, alternatives, and a recovery path when the email never arrives.

Primary action

agent.primaryAction tells the agent whether to enter an OTP or open a magic link.

Safe HTML buttons

primaryButton exposes buttons like “Verify email” or “Reset password” without scraping raw HTML.

Confidence metadata

Responses include confidence, matchedRule, reason, and alternatives for safer automation.

Diagnose on timeout

diagnose returns failure summary, recommended action, and retry payloads for agents and CI logs.

Run timeline

Trace inbox created, wait started, message received, extraction, callback, notify, and cleanup events by runId.

Cleanup policies

Delete successful inboxes, keep failures for debugging, or clean a whole CI run by labelPrefix.

Short-lived run keys

mailagent_issue_access mints temporary scoped keys for a single agent run. Guide →

Raw MIME + attachments

Full .eml and attachments are available when a preview is not enough. Guide →

MCP OAuth

Browser login for remote MCP — no API key in client config. OIDC guide →

Hosted console

Dashboard for inboxes, domains, team keys, and audit log. Open console →

Workspace Agent

Summarize threads, draft replies, suggest reminders, and execute sends with draft_only or allowlisted auto-send. Autonomy guide →

Forwarded mail workflow

Forward or send mail into a MailAgent inbox. Workspace tools summarize stored messages and draft replies without Gmail API access.

For QA teams

Stop fighting shared test inboxes

Each Playwright worker gets its own address. Filter by label, narrow with subjectContains, keep failed inboxes for debugging, and simulate email without SMTP in CI. Playwright guide → · MCP signup QA →

Parallel runs collide

label=pw-0-… per worker — list and debug inboxes for one CI job without stepping on teammates.

Wrong email in the pile

Service presets, subjectContains, and messageIndex keep the test on the intended email.

Leaked CI secrets

Scoped keys with labelPrefix=ci- limit blast radius if a pipeline secret escapes. Scoped keys →

              Playwright
              open (one-shot)
            

              import { createMailAgentQa, MailAgentQa } from "@mailagent/qa";

const mail = createMailAgentQa();
const { address, id } = await mail.createInbox({
  label: MailAgentQa.runLabel("signup"),
  service: "auth0",
});

// … fill signup form with address …

const { otp } = await mail.waitForVerification(id, {
  subjectContains: "verify",
});
            

Read the QA guide — labels, callbacks, CI env, Playwright flow. Scoped keys for CI →

Workspace Agent

From verification to mail you can act on

After signup QA, agents keep working the thread: summarize context, draft a reply, log actions, and complete reminders. Autonomous sends stay off by default; admins set policy, idempotency keys prevent double-send, and mailagent_workspace_model_status blocks execution when no LLM is configured.

Summarize & draft

mailagent_workspace_summarize and mailagent_workspace_draft_reply load mail from an inbox id or supplied message payload. DeepSeek and Qwen fall back to each other when both keys are set.

Reminders & audit trail

Suggest, create, and complete reminders. Log every agent step with mailagent_workspace_list_actions so open reminders are not drafted twice.

Policy-gated send

mailagent_workspace_execute_reply dry-runs first, then sends at most once per idempotency key. Modes: draft_only, auto_send_safe, or full_auto with domain allowlists.

Local LLM (dev)

WORKSPACE_LLM_PROVIDER=local with Ollama or LM Studio on the same machine as npm run dev — or use npm run dev:docker:ollama for a one-command stack. Local LLM guide →

              MCP
            

              // Check LLM before workspace work
await mailagent_workspace_model_status();

// Draft from stored inbound message
await mailagent_workspace_draft_reply({
  inboxId: "inb_…",
  messageId: "msg_…",
  instruction: "Confirm QA review — no new commitments.",
});

// Send only when policy allows + idempotency key
await mailagent_workspace_execute_reply({
  inboxId: "inb_…",
  messageId: "msg_…",
  idempotencyKey: "run-4821-reply",
});
            

MCP

MCP tools. Verify, workspace, simulate

Add mailagent to Cursor or Codex. Signup flows use structured primaryAction; workspace tools summarize MailAgent inbox messages, track reminders, and execute policy-gated replies. Mint scoped run keys and call mailagent_plan_next for recovery when a run stalls.

Tool	Description	Category
mailagent_workspace_model_status	LLM readiness (DeepSeek, Qwen, local) — call before workspace execution	Workspace
mailagent_workspace_draft_reply	Draft from stored MailAgent inbox message or supplied payload	Workspace
mailagent_workspace_execute_reply	Policy-checked send with idempotency — default is deny	Workspace
mailagent_simulate_message	Inject realistic test email without SMTP	QA
mailagent_issue_access	Short-lived scoped key for one autonomous run	Runs
mailagent_plan_next	Next MCP tool, payload, and recovery steps	Autopilot
mailagent_verify_signup	Wait → extract → return agent.primaryAction	Recommended
mailagent_create_inbox	Temporary address with service preset, TTL, notifyEmail, cleanup	—
mailagent_simulate_message	OTP, magic_link, invite, attachments — no SMTP in CI	—
mailagent_diagnose_inbox	Explain timeouts and return retry payloads	—

Full tool list: GET /v1/agent · Agent docs

Positioning

Agents, not burner extensions

Consumer burner mail protects your personal inbox. MailAgent automates verification for code.

	MailAgent	Mailosaur / QA SaaS	Typical burner mail
MCP / API-first	Yes	REST	Browser UI
OTP + link extraction	Built-in	Built-in	Manual
SSE wait	Yes	Poll	—
Per-run isolated inbox	label + API	Yes	Shared pool
Sender allowlist	Yes	Patterns	Block in UI
Self-hosted domain	Your Resend	Their domain	Their domain
Workspace summarize / draft	MCP + REST	—	—
Google Gmail OAuth	Disabled	Varies	—
Policy-gated auto-reply	Opt-in	—	—
Local LLM (Ollama)	Self-host dev	—	—
Long-lived aliases	TTL only	Yes	Yes

Open source

Hosted API when you need speed. Self-host when you need control.

MailAgent ships as a complete stack: Worker, MCP server, QA SDK, docs, and deploy scripts. Use our API now, or run the same architecture on your Cloudflare account.

View on GitHub Integration guide

License MIT free to use, fork, and self-host

Production stack Workers, Queues, Durable Objects, Neon the same pieces used by the hosted service

Hosted endpoint api.webmailagent.com optional API key for teams that do not want to operate infra

API

The response is already agent-readable

              curl
            

              # One-shot (recommended)
curl -X POST https://api.webmailagent.com/v1/inboxes/open \
  -H "Authorization: Bearer $API_KEY" \
  -d '{"service":"github","timeoutSeconds":90}'

# → verification.otp
# → verification.primaryLink
# → verification.primaryButton
# → verification.confidence
            

Full reference: API docs · OpenAPI

Give your agent
an inbox & a workspace

Signup QA: disposable inboxes, OTP extraction, CI simulate. Workspace: summarize, draft, reminders, policy-gated send. Open source — self-host or use the hosted API.

Star on GitHub API docs Console Hosted API key

Email verification for CI and AI agents

Verify signup and workspace mail

Verify signup in one call

MCP for Cursor & Codex

Playwright & Cypress

Self-host or hosted

Scoped run keys

No SaaS markupfor CI verification

Up and running in 1, 2, 3…

Create inbox

Wait & extract

Ship in CI

Return the action, not raw email markup

Primary action

Safe HTML buttons

Confidence metadata

Diagnose on timeout

Run timeline

Cleanup policies

Short-lived run keys

Raw MIME + attachments

MCP OAuth

Hosted console

Workspace Agent

Forwarded mail workflow

Stop fighting shared test inboxes

Parallel runs collide

Wrong email in the pile

Leaked CI secrets

From verification to mail you can act on

Summarize & draft

Reminders & audit trail

Policy-gated send

Local LLM (dev)

MCP tools. Verify, workspace, simulate

Agents, not burner extensions

Hosted API when you need speed. Self-host when you need control.

The response is already agent-readable

Give your agentan inbox & a workspace

No SaaS markup
for CI verification

Give your agent
an inbox & a workspace