Turn Your AI Agent Into an X Research Assistant: 5 Jobs

Give your AI agent X (Twitter) data with one MCP URL or API key, then run 5 research jobs: copyable prompts, real output and what each costs a month.

Tarun Khatri
Published 25 October 2026 · Updated 25 October 2026 · 7 min read · 阅读中文版

To turn your AI agent into an X research assistant, connect it to a source of X data, then give it five standing jobs.

Muse, Claude, ChatGPT, Codex, Cursor and OpenClaw can't read X (Twitter) on their own. This guide connects yours through Weir, which we run. Then you get five jobs to paste, each with a real call from 25 October 2026 and its monthly cost.

Add this URL to any MCP client, then ask your agent "What's my Weir usage?":

https://api.weirapp.io/mcp

How to give your AI agent access to Twitter

Your agent needs a tool that returns X posts as data. A pasted link won't do: in our test, x.com sent AI fetchers a 402 error instead of the post.

X's own MCP server needs your own X developer app, per X's docs, and the X API behind it is pay-per-use. Weir is one public URL instead:

Your agent	How it connects	What you give it
Claude, ChatGPT, Cursor	Sign in (OAuth)	The MCP URL
Claude Code, Codex, OpenClaw	Sign in or key	The MCP URL
Muse	Custom connector	A setup link and a key
n8n	MCP Client Tool	The URL and a Bearer key
Coze	MCP plugin	The URL and a header key
Python or curl	REST API	An Authorization header

If your agent signs in, you need no key. Otherwise, create one under API keys in the dashboard, with only the X scope and an expiry.

The Muse and Coze rows follow their own docs; we haven't tested them.

This script runs job 3 on Python's standard library, for sandboxes with no pip:

import json, os, urllib.request

body = {"fromUsers": ["OpenAI", "github"], "query": "-filter:nativeretweets",
        "onlyOriginal": True, "limit": 5}
req = urllib.request.Request(
    "https://api.weirapp.io/v1/x/search", data=json.dumps(body).encode(),
    headers={"Authorization": "Bearer " + os.environ["WEIR_API_KEY"],
             "Content-Type": "application/json"})
with urllib.request.urlopen(req, timeout=60) as r:
    print("credits:", r.headers["x-credits-charged"])
    for post in json.load(r)["data"]:
        print(post["likeCount"], post["url"])

Muse custom connector: add Weir to Meta's agent

Muse is Meta's personal AI agent, available in the US and Canada, says Meta's newsroom. Its Help Center says you can ask Muse to create a Custom Connector for a service it doesn't list. Paste this into Muse:

Create a custom connector for Weir by following https://docs.weirapp.io/muse.md. Use my Weir API key as a Bearer token.

Our Meta Muse guide has the full steps and priced prompts.

Keys stay out of the chat. Meta says Muse keeps them in its Secure Credentials Store.
Meta doesn't review custom connectors. It tells you to check each provider's privacy policy first.
We haven't tested these screens. The steps follow Meta's Help Center and a guide by Parallel, so labels may differ.

Job 1: Follow a hot topic by keyword and time window

A keyword and a time window catch a story while it moves:

Use Weir's search_x: query "Brotli", last 48 hours, 50+ likes, English,
no replies, limit 5. Per post: likes, one line on what it says, link.
Flag claims you can't check.

We ran it on 25 October 2026, two days after OpenAI said its agents had found a bug in Brotli. All five posts came from personal accounts, so we show counts only:

→ search_x {"query": "Brotli", "since": "2026-10-23T08:10:00Z", "minLikes": 50,
            "onlyOriginal": true, "language": "en", "limit": 5}
← 5 posts · 6 credits (1 + 5 × 1)
  119 likes · retells OpenAI's finding
   50 likes · a severity score and a version list
   62 likes · says one Rust port is safe, with a link
  100 likes · retells OpenAI's finding
  114 likes · a reaction, no new facts

Two retell the same news, which job 5 merges for you.

One gives numbers you should check at the source before you act.

Monthly cost: one 20-post page a day comes to about 630 credits.

Job 2: Spot deals and free credits

Search an exact phrase, then tell the agent what doesn't count:

Use Weir's search_x: phrase "free credits", last 7 days, 100+ likes,
English, no replies, limit 5. Keep offers made by the company itself;
drop invite codes and "comment to win" posts. Per offer: who, what,
deadline, link.

Our run returned five posts for 6 credits, and all five were promotions. Three asked you to comment, repost or use an invite code.

The one brand account, @CMG_Esports, offered "a chance at free credits" for a retweet and a comment.

That's why the prompt filters before it lists.

Monthly cost: daily, about 630 credits.

Job 3: Watch specific accounts

Name the accounts once. X search includes reposts by default, so -filter:nativeretweets drops them, and onlyOriginal drops replies:

Use Weir's search_x: fromUsers ["OpenAI", "AnthropicAI", "XDevelopers",
"github", "cursor_ai"], last 48 hours, query "-filter:nativeretweets",
onlyOriginal true, limit 5. Per post: account, date, news, likes, link.

← 5 posts · 6 credits · newest first
  25 Oct  @github  Universe badge that "runs Doom"      127 likes
  24 Oct  @github  Evergreen app for collaborators       142 likes
  23 Oct  @OpenAI  Brotli flaw from red-teaming       6,602 likes
  23 Oct  @github  "While you're waiting for GTA 6..."    921 likes
  23 Oct  @OpenAI  training security update          13,090 likes

The five newest came from two of the accounts. Raise limit to see more.

Monthly cost: daily, about 180 credits for 5 posts a day. For one account, the free tier's monitor can check daily and bill only new posts.

Job 4: Read real community feedback, posts and replies

Replies under a launch post show what users think:

Use Weir's get_x_replies on https://x.com/OpenAI/status/2113718797505036665,
sort "likes", one page. Group the replies into themes, count each, and
give one example link per theme. List the questions people asked.

X returned 34 replies for 35 credits. We grouped them by hand, as your agent would:

34 most-liked replies to OpenAI's 23 October post

Doubts about the security story9

Jokes and memes8

AI safety worries5

An image or GIF only5

Product asks and complaints5

Other2

Sorting by likes brings jokes to the top: 13 of the 34 were jokes or bare images.

The real feedback sat in the doubts and the product asks.

Monthly cost: two reply pages a week, about 280 credits. Reddit opinions cost 3 credits a post with search_reddit.

Job 5: Have the agent dedupe, group and summarize

This job makes no new calls. It turns the posts you already paid for into a brief:

From the posts above: merge posts about the same story, group the rest,
and write at most 5 lines, each with its link. Name claims you couldn't
check. End with the credits these calls used.

We wrote this brief from our four runs; your agent's wording will differ:

X brief · 25 Oct 2026 · 53 credits
1. OpenAI tightened training security; training resumed on 1 Oct.
   https://x.com/OpenAI/status/2113718797505036665
2. OpenAI's agents found a Brotli flaw (2 retellings merged).
   https://x.com/OpenAI/status/2113766354386084338
3. GitHub's Universe badge runs Doom.
   https://x.com/github/status/2114145673033691289
Unchecked: one post's Brotli severity score.
Replies: 9 of 34 doubted OpenAI; 5 asked for products.

What does an X research assistant cost a month?

All five jobs use about a third of your free credits:

Credits a month for the five jobs, 1,720 in all

Hot topics, daily630

Deals, daily630

Replies, twice a week280

Accounts, daily180

The brief0

Those runs read about 1,622 posts. On X's own API, the same posts would cost you about $8.

Failed calls cost nothing, and every call reports its cost, as meta.creditsCharged in MCP results and the x-credits-charged header over REST, so your agent can add up its bill.

When not to use Weir

You need posts within seconds. Weir answers on request or on a schedule; it doesn't stream.
Your agent should post or reply. Weir is read-only, while X's API and TwitterAPI.io can write.
Cost per tweet is all that matters. TwitterAPI.io lists $0.15 per 1,000 tweets, as of October 2026.
You need private data, like your bookmarks or DMs. Weir reads public posts only.

Weir isn't licensed by X, so access can change. Check X's terms; this isn't legal advice.

FAQ

Can an AI agent read Twitter for free?

Yes, within limits. Weir gives you 5,000 free credits a month, about 4,700 X posts, with no card. X's own API has no free tier; our free Twitter API guide compares the $0 options.

Can Meta's Muse read X posts?

Through a custom connector, yes: Muse builds one from Weir's setup page, as our Meta Muse guide shows. We haven't tested the Muse screens.

Can my agent post on X through Weir?

No. Weir's tools only read public data: none can post, like, follow or send DMs.

Sources

All checked on 25 October 2026:

Meta Help Center, how Muse works with Connectors: https://www.meta.com/help/artificial-intelligence/1687253048996149/
Meta Newsroom, Muse for Small Business (29 September 2026): https://about.fb.com/news/2026/09/introducing-muse-small-business/
Parallel, custom integrations with Meta Muse: https://parallel.ai/articles/meta-muse-custom-integrations
X docs, MCP servers and API pricing: https://docs.x.com/tools/mcp and https://docs.x.com/x-api/getting-started/pricing
Coze docs (in Chinese), a plugin from an MCP service: https://docs.coze.cn/guides_create_a_plugin_based_on_mcp.md
TwitterAPI.io pricing and posting: https://twitterapi.io/pricing and https://docs.twitterapi.io/api-reference/endpoint/create_tweet_v2
OpenAI on X, 23 October 2026: https://x.com/OpenAI/status/2113718797505036665 and https://x.com/OpenAI/status/2113766354386084338
Weir docs: https://docs.weirapp.io/mcp and https://docs.weirapp.io/billing
Our four Weir calls, 25 October 2026: 53 credits