Keep a Manus Cue Connector for Lead Finding Out of Strangers' Inboxes

Every Cue agent comes with its own email address and phone number, which makes a Manus Cue connector for lead finding the one setup where the obvious shortcut is the wrong one. Cue is the personal agent app Manus launched on 28 September 2026. Hand one of its agents a list of people asking for a product like yours, and it holds everything it needs to write to them. Don't let it.

The short version: connect RedReplier as a custom MCP connector, let the agent read and explain HIGH and VERY_HIGH mentions, and write into its brief that it never contacts anyone it finds. RedReplier drafts replies and never posts them. The agent's own outbox is the part you close.

Why not let a Cue agent message leads itself?

Letting a Cue agent message leads puts a note from an agent identity in front of a person who never asked to hear from you. Manus describes each Cue agent as having "its own email, phone number, wallet, and computer, so it can send messages." That capability exists for errands. A cold message to a stranger found in a public thread is not an errand.

The failure stacks up fast. The recipient gets an unsolicited email or text from an address they cannot trace to a human, which reads as spam whatever it says. Any complaint lands on the agent's address, which you do not read. And someone who asked a question in public on Reddit expects an answer in the thread, where other readers judge it, not a private message that proves you went looking for them. Our notes on Reddit's promotion rules cover what subreddits already think of vendors who show up uninvited.

Manus has not documented a way to stop this at the platform level. Runtimewire noted on 28 September 2026 that the Cue announcement "does not explain in detail how users approve individual transactions, restrict what agents can do with their email or phone identity, or review actions after they occur." As of 29 September 2026 the restriction lives in your brief or nowhere.

What does RedReplier do with a reply?

RedReplier writes a draft reply for scored mentions and never publishes one. The spec describes the aiReplySuggestion field as a "draft reply for a human to review before posting," and none of RedReplier's 22 API operations posts to Reddit, X, Bluesky, Hacker News or Facebook. The connector gives Cue a draft to show you and no way to send it through RedReplier.

That is the gap. A draft in the agent's hands plus an email address in the agent's hands makes a sent message, unless the brief forbids it. Leave drafts out of the digest unless you ask for one, because the step from reading suggested wording to sending it is short. When you do answer, answer as yourself in the thread where the question was asked. RedReplier's Reddit lead generation page covers that workflow without an agent.

What do I paste into Cue for a lead digest?

A lead digest brief names the allowed tools, the filters, and the contact ban in one message. The connector itself goes in first, and the Manus Cue connector for Reddit monitoring pillar covers the Manus settings path, the Editor-role token and the test for whether Cue sees the server.

You have a RedReplier connector. Your job is a lead shortlist for me.
 
Allowed tools: list_mentions, count_mentions, explain_mention.
Never call update_mention_status, delete_keyword, add_keywords,
edit_keyword or update_website.
 
Every weekday morning, list NEW mentions found in the last 24 hours,
HIGH and VERY_HIGH only, sorted by RELEVANCE. For each one: source,
score, one line on what the person is trying to buy, and the link.
Explain the top three. Do not show me aiReplySuggestion unless I ask.
 
You never contact anyone who appears in a mention. Not by email,
text, phone call, DM or reply, and not by looking them up elsewhere.
If a mention has no link, say so and stop there.
Report to me in this chat only.

The last paragraph does two jobs. It bans contact through the agent's own identity, and it bans the side route: an agent that cannot find a link for a mention goes looking for the person elsewhere, which is the first step of outreach. RedReplier returns a null url when the platform gives no public link, which the spec says happens with some Facebook and Bluesky posts.

How do I split lead work across a Cue group chat?

A Cue group chat splits lead work well when one agent reads, one agent researches, and only you reply. Manus says agents in a Cue group chat "hand work to each other," and adds: "You set the direction and make the final call." For lead finding, that last sentence is the whole design.

Give the RedReplier connector to one scout agent with the brief above. A second agent takes the shortlist, reads each linked thread, and tells you which ones are a fit and why. Neither agent's brief says "follow up." You are the only member of the chat who writes to anyone outside it.

As of 29 September 2026 Manus has published nothing on isolation between agents, or on whether agents in a group chat share connectors. Test it: ask the research agent to list your RedReplier websites. If it answers, it holds the connector too, and its brief needs the same contact ban.

Which RedReplier calls must stay off a lead finding agent?

Four RedReplier calls do damage specific to lead finding, and update_mention_status leads the list. It approves or rejects a mention, and a rejected mention leaves the NEW queue, so a lead the agent misjudged disappears without a trace.

delete_keyword deserves its own sentence: RedReplier's agent docs say it deletes "a keyword and the mentions it found," so a lead history goes with it. add_keywords makes keywords ACTIVE at once when the plan allowance covers them. An agent that decides your market is wider than you think starts matching threads you never chose. Asking the agent to suggest keywords is fine. Letting it add them is not.

How do I find leads rather than mentions?

RedReplier finds leads when the filters favour buying questions over brand chatter. Four filters on GET /mentions do the work: scoreBuckets limits the list to HIGH and VERY_HIGH, sources narrows to REDDIT_COMMENT where recommendation threads live, keywords limits to competitor names, and includeLowRelevance stays at its default of false.

Then read the reasoning, not the number. explain_mention fills in relevanceReason, which tells you whether the person is asking for a recommendation or complaining about a tool they already chose. Our buyer intent write-up covers which signals justify a reply.

What breaks on lead finding?

The digest window breaks first. "Since the last run" is the agent's own bookkeeping, not an API field, so a missed morning either repeats yesterday or skips it. The from filter works on ingestedAt, so ask for a fixed 24 hours and accept a little overlap.

The draft arriving unasked breaks second. GET /mentions returns aiReplySuggestion with each mention, so an agent summarising the response hands you ready wording unless the brief names the field.

The token breaks third. Every RedReplier call needs redreplier.write, so a Viewer key returns 403 permission_denied on the first read. Create the key under the Editor role with an expiry date.

Frequently Asked Questions

Can Cue send a lead the RedReplier draft reply?

Cue has the means, because each agent has its own email and phone number, and nothing in RedReplier stops a message that never touches RedReplier. The contact ban in your brief is the control. As of 29 September 2026 Manus has not documented a setting that restricts an agent's email or phone identity.

Does RedReplier post replies for me?

RedReplier drafts replies and never posts them. The aiReplySuggestion field is a "draft reply for a human to review before posting," and none of the 22 API operations publishes anything. You post the reply yourself, in the thread.

Can a Cue agent spend money through RedReplier?

A Cue agent has a wallet that pays "within the budget you set," but the RedReplier API gives it nothing to pay for. POST /keywords/activate-pending never charges, and keywords beyond the plan stay PENDING until someone upgrades in the RedReplier app. The billing preview calls only report prices.

How is this different from the Muse and Grok versions?

The Meta Muse connector for lead finding builds a connector from the REST spec, and the Grok bot template for lead finder ships a ready brief. Meta says Muse asks for approval by default before it sends an email. For Cue, Manus documents no approval step and gives each agent its own email and phone number. That outbox is why Cue's version needs the contact ban.

Which sources count for lead finding?

GET /mentions covers REDDIT_POST, REDDIT_COMMENT, TWITTER, BLUESKY, HACKERNEWS, FACEBOOK and FACEBOOK_GROUP. Start with REDDIT_COMMENT, where people ask each other what to buy, then widen the sources filter. The social listening guide covers running all five platforms through one agent.

How do I shut the lead agent down fast?

Revoke its token at redreplier.com/api-tokens, and every RedReplier call fails from that moment. Then remove the connector in Manus, which Manus says clears "all login information and saved credentials." Revoking the token does not stop the agent's email or phone, so remove the lead brief from the agent too.

What should a Cue lead digest contain?

For each mention, the digest should show the source, the score, one line on what the person is trying to buy, and the link. Limit it to HIGH and VERY_HIGH mentions from the last 24 hours, sorted by RELEVANCE. Ask the agent to explain the top three and to leave aiReplySuggestion out unless you request it.

Which RedReplier token role does a Cue lead agent need?

Create the key under the Editor role and give it an expiry date. Every RedReplier call needs redreplier.write, so a Viewer key returns 403 permission_denied on the first read.

What should the agent do when a mention has no link?

It should say so and stop. RedReplier returns a null url when the platform gives no public link, which the spec says happens with some Facebook and Bluesky posts. Looking for the person elsewhere is the first step of outreach, so the brief bans it.

Why does my lead digest repeat or skip mentions?

"Since the last run" is the agent's own bookkeeping, not an API field, so a missed morning either repeats yesterday or skips it. The from filter works on ingestedAt. Ask for a fixed 24 hours and accept a little overlap.

Create a RedReplier API token with the Editor role and an expiry date, give one Cue agent the lead brief above, and read tomorrow's shortlist yourself before anyone replies to anything.

Sources: Manus's Introducing Manus 2.0 (28 September 2026), How can I use Manus Connectors, the custom MCP docs, Runtimewire (28 September 2026), the RedReplier OpenAPI spec and agent docs. Checked 29 September 2026.