What it does
The human-approval queue: AI-written replies waiting for someone to press send. An agent can read and summarise them; it cannot approve them.
Arguments
None. Send an empty arguments object.
This table is generated from the literal inputSchema that tools/list returns, so what your client sees and what this page says are the same object:
{
"type": "object",
"properties": {}
}Request
Paste this, swap the key, and you have called the tool.
curl -s https://topclozer.com/api/mcp \
-H "Authorization: Bearer tcz_live_YOUR_KEY" \
-H "Content-Type: application/json" \
-d '{
"jsonrpc": "2.0",
"id": 1,
"method": "tools/call",
"params": {
"name": "list_tasks",
"arguments": { }
}
}'Returns
An array of tasks: id, type (always approve_draft today), campaignId, conversationId, messageId, createdAt. Join messageId against get_inbox to read the draft text.
The payload arrives inside an MCP content block — one text block whose body is this JSON, pretty-printed. Parse the string; there is no structured-content field.
[
{
"id": "tsk_9d2",
"type": "approve_draft",
"campaignId": "cmp_7f21a0",
"conversationId": "cnv_51ab",
"messageId": "msg_3",
"createdAt": "2026-09-03"
}
]Worth knowing
The task object carries ids, not text. Approval stays in the app on purpose: sending a message is the one irreversible action in the product, so it is not exposed to an agent.
Limits
No request limit. This tool reads the workspace database and costs nothing to serve. The limits that bite in this product are the sending controls — per-sender weekly invite caps, the warm-up ramp, the working-hours window and a randomised gap between actions — and no tool can raise them.
Related tools
All 21 tools are listed on the tool index. Connecting a client is on the Claude setup page, and the protocol details are on the server page.