Document intake agent: read the invoices, quotes and spreadsheets that arrive as attachments
ThreadCamp gives the agent a real email address so it can read the invoices, quotes and spreadsheets that arrive as attachments - with programmatic inboxes, wait_for_message for time-sensitive replies, and human approval when a send matters.
Give the agent an inbox
Call POST /v1/inboxes (or create_inbox over MCP) to get a working address in milliseconds - on the shared relay.threadcamp.com domain or your own custom domain. The agent reads inbound mail, replies on the thread and forwards, all with the same key.
The file, not just the covering note
Most of what matters arrives attached rather than typed. GET /v1/attachments?thread_id=... lists what is on a conversation - filename, type and size - and GET /v1/attachments/:id mints a short-lived signed URL the agent fetches directly. The bytes never pass through the API, so a forty-megabyte spreadsheet costs nothing to hand over.
A body that says nothing is not an empty message
"Please see attached" is indistinguishable from an empty reply unless something can enumerate the files. Listing is what makes that visible, which is why an agent should call it before deciding a message contains no information.
Wait for the message that unblocks the task
Many agent jobs stall on an email arriving - an OTP, a verification link, a reply. wait_for_message polls the inbox for the specific message the agent needs mid-task, so it can continue without a brittle sleep-and-retry loop.
// Poll a ThreadCamp inbox for an incoming OTP or verification email
const msg = await fetch("https://www.threadcamp.com/v1/inboxes/inbox_.../wait", {
method: "POST",
headers: { Authorization: "Bearer sk_live_...", "Content-Type": "application/json" },
body: JSON.stringify({ from: "no-reply@service.com", timeout_s: 120 }),
}).then((r) => r.json());
// -> { "id": "msg_...", "text": "Your code is 481920", ... }Keep a human in the loop
For anything that sends outward, set the inbox to approve-first. Every agent send is held as a pending draft in a human approval queue; a one-click approval is what actually executes the send. That is how a document intake agent runs unattended without becoming a spam risk.
Related jobs
OTP retrieval agent
retrieve one-time passcodes and 2FA codes mid-task
Support agent
hold email conversations with customers, with human approval on replies
Signup automation
sign up for services that require email verification
Powered by the email platform for AI agents and human-in-the-loop approvals.
Questions
How does ThreadCamp support a document intake agent?
Create a real inbox with POST /v1/inboxes, then let the agent read the invoices, quotes and spreadsheets that arrive as attachments. It reads inbound mail over signed webhooks or polling, uses wait_for_message for time-sensitive replies like OTPs, and can be set to approve-first so a human signs off before anything sends.
Is a document intake agent safe to run unattended?
Each inbox has an autonomy policy - auto, approve-first or notify. Under approve-first every send lands as a pending draft in a human approval queue. Scoped API keys, sandbox mode, rate limits and bounce auto-pause bound the blast radius.
Build your document intake agent today.
No credit card. Full API, MCP and webhooks on the free tier.