Set Up a Meta Muse Connector for X Social Listening

As of 20 September 2026, X appears on no published connector list, which makes a Meta Muse connector for X social listening something you build rather than something you install. Meta launched Muse on 8 September 2026 with Facebook, Instagram and Threads wired in automatically through Accounts Centre. The roster Chief AI Officer Alexandr Wang posted on launch day runs from Gmail and Google Calendar to Spotify, Plaid and Peloton, and X is not on it.

The short version: ask Muse to build a custom connector against RedReplier's OpenAPI spec at https://redreplier.com/openapi.json, paste a redreplier_ token into the secure credential prompt rather than the chat, and ask for sources=TWITTER sorted by relevance. Meta does not review custom connectors or how they use your information.

Why is there no built-in X connector for Muse?

As of 20 September 2026, X appears on no Muse connector list Meta or its executives have published, and Meta publishes no connector directory outside the product. Threads is Meta's short-post network, and Meta connects Threads, Instagram and Facebook to Muse automatically when the accounts sit in the same Accounts Centre. The networks Meta wired in first are the ones it owns.

Meta has not said whether an X connector is planned. As of 20 September 2026 it has announced none, so the custom connector is the only route that exists today.

What Meta does document is the alternative. If a service is not in the Connector list, you ask Muse to build a Custom Connector, and Muse writes the client itself on the per-user virtual machine it runs on in Meta's cloud. The Meta Muse connector for Reddit monitoring guide covers the no-review warning, the credential store and why a server on your own machine is unreachable. One point carries straight over to X: Parallel, testing custom connectors on 14 September 2026, reported that Muse has no add MCP server setting, and no Meta help centre article for the consumer agent mentions MCP.

What does X listening actually need?

X listening needs deduplication, scoring and a status per mention, because volume is the entire problem on X. A keyword that returns nine Reddit threads a day returns hundreds of posts on X, most of them one line long, and the same thought repeated by different accounts.

RedReplier scores every X mention from 0 to 100 before the connector sees it, so the agent reads a shortlist instead of a feed. GET /mentions returns that relevanceScore alongside a status of NEW, APPROVED or REJECTED, which is what stops this morning's digest from re-showing what you dealt with yesterday. Our guide to X social listening covers the same triage problem without an agent in the loop.

The other half is cross-source coverage. sources is an array, so GET /mentions returns the Reddit, Hacker News and Bluesky versions of the same conversation in the same response as the X posts, and the long version arrives with the one-line version.

What changes in the setup message when the source is X?

The paste block is the pillar's, with the last paragraph swapped. Everything about the spec URL, the bearer header and the credential prompt is in the Meta Muse connector for Reddit monitoring guide. The X-specific lines are these:

Use GET endpoints only. Do not call anything that creates, updates or
deletes, and never change a mention status. Start with today's X
mentions: sources=TWITTER, sort=RELEVANCE, statuses=NEW,
scoreBuckets=HIGH, scoreBuckets=VERY_HIGH.

How does relevance scoring cut X volume?

MentionSort takes RELEVANCE or RECENT, and on X that setting decides whether the agent hands you a digest or a raw transcript. Recency ordering surfaces whatever posted most recently, which on a high-volume source returns the last ten minutes rather than the day's best threads.

Pair the sort with scoreBuckets and the list drops to what deserves a human minute. Ask for HIGH and VERY_HIGH only, then ask Muse to call POST /mentions/{id}/explain on the top two. That call generates and stores the reasoning behind a score, which is why it is a POST rather than a GET, and an agent told "GET only" will skip it unless you name it.

There is a Muse-side reason to filter hard as well. Muse meters usage in Muse tokens on a weekly allowance. As of 20 September 2026 Muse is free up to that limit, with a Power plan at USD 20 a month for 500 million Muse tokens a week and a Max plan at USD 100 a month for 3 billion. Parallel reports that custom connector work draws on that same weekly meter rather than a separate one, so every mention the connector pulls into the conversation is tokens spent reading X posts. A 400-item response on a broad keyword is an expensive way to find four things.

Why go through an aggregator instead of holding an X API tier yourself?

Holding your own X API access means paying X per post read and building the scoring layer yourself. As of the 8 September 2026 check on our X API pricing page, X bills pay-per-use credits at $0.005 per post read with a cap of 3 million post reads per monthly cycle, the free tier is closed to new developers, and the legacy $200 Basic and $5,000 Pro plans stay open to existing subscribers only. Enterprise access starts at $42,000 a month.

X's API returns posts. RedReplier returns the same posts with a relevanceScore, a status of NEW, APPROVED or REJECTED, and a reviewedAt timestamp, which is the part an agent reads instead of rebuilds. Those are the parts an agent needs, and on your own X tier they are the parts you write yourself.

What breaks on X specifically?

Two things go wrong on X that barely register on Reddit. Leaving sort unset is the first, because RedReplier's spec declares no default order for GET /mentions, so the digest arrives in whatever order the API returns. Name sort=RELEVANCE in the setup message. Broad keywords are the second: a product name that doubles as a common word matches enormous X volume and drags the score distribution down with it, so HIGH and above returns a list that is still too long.

Two failures that look X-specific are not. The browser fallback and the pending-keyword check behave the same on every source, and the pillar covers both.

Frequently Asked Questions

Does the connector need X API access of its own?

No. RedReplier collects the X mentions and the connector reads them through the RedReplier API with your redreplier_ token. There is no separate X developer account and no per-read bill; the only limit to pace against is RedReplier's own per-key rate limit, which every response reports in its RateLimit-* headers.

Can Muse post a reply to X for me?

No. RedReplier's public API has no endpoint that publishes a reply, so the connector drafts and surfaces rather than posts. A reply written in your name on X is a decision, and the API keeps that decision on your side of the line.

Why does the same X mention keep appearing?

Its status is still NEW. Marking it APPROVED or REJECTED through PATCH /mentions/{id}/status removes it from the new queue. Keep that call out of the connector's scope for the first week, because a rejected mention drops out of the statuses=NEW digest and only comes back when you ask for statuses=REJECTED.

Can Muse read my X mentions over MCP instead?

Parallel tested this on 14 September 2026 and reported that the Muse consumer agent has no add MCP server setting, so the custom connector is the route. No Meta help centre article for the consumer agent mentions MCP. RedReplier's hosted MCP server at https://mcp.redreplier.com/mcp exposes the same X mentions to Claude, Cursor and Codex behind the same token.

Which other agents can read the same X mentions?

Any client that speaks REST, MCP or plain markdown reads the same operations behind the same redreplier_ token. The Grok bot template for X social listening covers the Grok Bot route, which needs no custom connector because the template ships the skill already written.

How much does a Meta Muse connector for X social listening cost?

As of 20 September 2026 Muse is free up to a weekly limit, with a Power plan at USD 20 a month for 500 million Muse tokens a week and a Max plan at USD 100 a month for 3 billion. Parallel reports that custom connector work draws on the same weekly meter. Filtering to HIGH and VERY_HIGH keeps each run from spending tokens on low-signal X posts.

Does Meta review a custom connector for X?

Meta does not review custom connectors or how they use your information. That leaves the token in your hands, so paste it into the secure credential prompt rather than the chat.

Can the same connector watch Reddit and Bluesky as well as X?

It can. sources is an array, so GET /mentions returns the Reddit, Hacker News and Bluesky versions of a conversation in the same response as the X posts. Set sources=TWITTER when you want X alone.

What should I let Muse do on the first run?

Read only. Tell it to use GET endpoints, start with GET /mentions/count, and keep PATCH /mentions/{id}/status out of scope for the first week. Name POST /mentions/{id}/explain if you want the reasoning behind the top two scores.

Why does Muse return my X mentions in a strange order?

RedReplier's spec declares no default order for GET /mentions, so an unset sort arrives in whatever order the API returns. Name sort=RELEVANCE in the setup message. RECENT is the other option, and on X it returns the last ten minutes rather than the day's best threads.

Create a RedReplier API token and give Muse a read-only X feed today, or read the agent docs first.

Sources: Meta's Muse connectors help article, Meta's safety engineering post, Parallel's custom integrations write-up (14 September 2026), and the RedReplier OpenAPI spec. Checked 20 September 2026.