Shortlist X Chatter With a Manus Cue Connector for X Social Listening

X posts are short, fast and repetitive, so a Manus Cue connector for X social listening needs a filter before it needs a feed. Cue is the personal agent app Manus launched on 28 September 2026, and its launch material shows no X integration. A custom MCP connector to RedReplier gives a Cue agent scored X mentions, and the scoring is what turns the volume into a shortlist.

The short version: add RedReplier's MCP server as a custom connector, ask for TWITTER mentions with a count first, sort by RELEVANCE, and keep only HIGH and VERY_HIGH. RedReplier needs no X developer account from you, and it posts nothing to X.

Why not have Cue watch X directly?

Cue has no documented X integration as of 29 September 2026, and the one X example Manus published belongs to Manus, not Cue. Manus's help centre describes it: "Set up an Automation that watches a product account on X. When a new update is posted, Manus drafts a summary... You still decide what goes out." That example follows one account. Social listening follows keywords across every account, which is a different job.

Going to X yourself means holding X API access. X bills pay-per-use credits at $0.005 per post read with a 3 million read cap per billing cycle, per our X API pricing breakdown, and the raw posts arrive with no score. A Cue agent also has its own computer, but Manus has not documented how that computer handles a site like x.com, so browsing is a route you test, not one you follow.

RedReplier's GET /mentions returns X posts under the source value TWITTER, each with a relevanceScore and a status of NEW, APPROVED or REJECTED. The agent reads a ranked list, and you never open an X billing account.

What goes in the Cue brief for X?

The X brief adds three things to the standard one: a count before the list, an explicit sort, and a hard cap. The connector setup itself, the Manus settings path, the Editor-role token and the test for whether Cue sees the server, sits in the Manus Cue connector for Reddit monitoring pillar.

You have a RedReplier connector. You watch X for me.
 
Allowed tools: list_websites, list_mentions, count_mentions,
explain_mention. Never create, edit, delete, enable, disable,
approve or reject anything. Never change alert settings.
 
Each morning:
1. count_mentions for NEW TWITTER mentions from the last 24 hours.
   Tell me the number first.
2. list_mentions, same window, sources=TWITTER, sort=RELEVANCE,
   HIGH and VERY_HIGH only, limit 10.
3. For each: the author, the score, one line on what they said,
   and the link. Explain the top two.
 
Never contact anyone in a mention. No email, no text, no call.
Never log in to X or any other site on my behalf.
If the count is zero, send nothing.

The count comes first because it tells you whether today is a digest or a flood before the agent spends effort reading. RedReplier's spec declares no default for sort, so naming RELEVANCE keeps the order from being whatever the API returns.

How does relevance scoring cut X volume?

RedReplier scores each X mention before the Cue agent sees it, so the agent filters by number instead of reading every post. list_mentions takes scoreBuckets for the five bands from VERY_LOW to VERY_HIGH, and minScore for a floor between 0 and 100. includeLowRelevance defaults to false, which already drops mentions the scorer judged unlikely to deserve a reply.

sort=RECENT is the setting that ruins X digests. On a busy keyword, recency returns the last hour's posts, not the day's best. sort=RELEVANCE puts a thoughtful complaint about a competitor above fifty passing mentions of your brand name.

The score alone does not tell you why. explain_mention fills in relevanceReason and returns it, so you can tell a buying question from a joke that happened to match your keyword. It also fills in aiReplySuggestion when that field is missing, which is the draft wording the next section is about.

What must a Cue agent never do on X?

A Cue agent must never reply on X, contact an X user, or hold your X login. Manus gives each Cue agent "its own email, phone number, wallet, and computer," and as of 29 September 2026 Manus has not documented any control over how an agent uses that identity. Runtimewire noted on 28 September 2026 that the announcement does not explain how users "restrict what agents can do with their email or phone identity."

RedReplier itself posts nothing. The spec describes aiReplySuggestion as a "draft reply for a human to review before posting," and no RedReplier operation publishes to X. The risk is the agent acting outside RedReplier: messaging the author through a contact address it found, or posting from an X session you left logged in on its computer.

Unreviewed replies on X carry a cost of their own. A brand that answers every mention within minutes turns into a reply guy, and X's own policy makes accounts "thought to be engaged in spammy behavior" ineligible for recommendation, which our note on the X shadowban quotes. Read the shortlist, pick the posts worth answering, and answer them yourself.

Can Cue watch X and Bluesky together?

One Cue agent watches X and Bluesky together, because sources on list_mentions takes a list of values in one call. Add BLUESKY next to TWITTER and the same brief returns both, scored on the same scale.

Two details change when Bluesky joins. The spec says url is null when the platform gives no public link, "as with some Facebook and Bluesky posts," so tell the agent to report a missing link rather than search for the author. And count per source, since X out-produces Bluesky and a shared cap of 10 fills with X. The social listening guide covers all five platforms through one agent.

What breaks on X specifically?

The first X-specific failure is a broad keyword. A product name that doubles as a common word matches a flood of unrelated posts, and even HIGH returns a list too long to read. Fix the keyword in RedReplier, not the filter in the brief.

The second is the same mention showing up every day. A mention stays in the NEW list until someone approves or rejects it, and the brief forbids the agent from doing that. The 24-hour window on ingestedAt keeps each digest to fresh finds, and you clear the queue yourself in RedReplier.

The third is the token. RedReplier checks redreplier.write on every call, reads included, so a Viewer key returns 403 permission_denied on the first count_mentions. Create the key under the Editor role at redreplier.com/api-tokens, with an expiry date.

Frequently Asked Questions

Does the Cue agent need X API access of its own?

The Cue agent needs no X API access, because RedReplier collects the X mentions and the connector reads them with your redreplier_ token. You hold no X developer account and pay X nothing per read. The limit to watch is RedReplier's per-key rate limit, reported in RateLimit-Remaining on every response.

Can Cue post a reply on X through RedReplier?

RedReplier has no endpoint that publishes to X, so the connector cannot post. Each mention carries an aiReplySuggestion draft for a person to review. Keep your X login off the agent's computer, because RedReplier's limits do not cover a session the agent holds itself.

Why does the same X mention keep appearing?

An X mention stays NEW until someone approves or rejects it through PATCH /mentions/{id}/status. Keep that call away from the agent, since a rejected mention drops out of the NEW list for good unless you ask for statuses=REJECTED. Filter each digest to the last 24 hours and triage in RedReplier yourself.

Can a Manus Automation watch X instead of Cue?

Manus documents an Automation that watches a product account on X and drafts a summary, with the note "You still decide what goes out." That covers one account, not keyword listening. A scheduled Manus Automation using the same RedReplier connector covers keywords, and the pillar explains that route.

How does this compare with the Muse and Grok versions?

The Meta Muse connector for X social listening writes a REST client from RedReplier's spec, and the Grok bot template for X social listening ships the brief ready to install. Cue takes an MCP server and gives the agent its own email and phone number. That identity is why the Cue brief carries a contact ban.

Is Cue open to everyone yet?

Cue is invite-only as of 29 September 2026, free during early access, and Manus has not said what it will cost afterwards. The code MEETCUE works for a limited number of early users, and each new user receives extra codes to share. RedReplier's social media monitoring page covers X listening without an agent in the meantime.

What X API cost does the connector avoid?

X bills pay-per-use credits at $0.005 per post read, with a 3 million read cap per billing cycle. Through the RedReplier connector you open no X billing account. The posts also arrive already scored, so the agent reads fewer of them.

Which score bands should a Cue agent read for X mentions?

Start with HIGH and VERY_HIGH, which are the top two of five bands running from VERY_LOW. Use minScore if you want a numeric floor between 0 and 100. includeLowRelevance defaults to false, so mentions the scorer judged unlikely to deserve a reply are already dropped.

Why does the first count_mentions call return 403?

RedReplier checks redreplier.write on every call, reads included, so a Viewer key returns 403 permission_denied. Create a new key under the Editor role at redreplier.com/api-tokens and give it an expiry date. Then rerun the count.

Should a Cue agent sort X mentions by recent or relevance?

Sort by RELEVANCE. On a busy keyword, sort=RECENT returns the last hour's posts instead of the day's best. RedReplier's spec declares no default for sort, so name it in the brief.

Create a RedReplier API token with the Editor role, connect it to one Cue agent, and ask for tomorrow's X count before a single post.

Sources: Manus's Introducing Manus 2.0 (28 September 2026), What is new in Manus 2.0, the custom MCP docs, Runtimewire and Implicator.ai (28 September 2026), the RedReplier OpenAPI spec and agent docs. Checked 29 September 2026.