MCP for Marketing: Research and Listening, Not Posting
MCP for marketing research: five prompts for Claude or ChatGPT that read public X (Twitter) posts, with the real output and credit cost of each one.
Tarun Khatri
Published · Updated · 7 min read
MCP for marketing lets Claude or ChatGPT read live data, and this page covers research and listening, not posting.
You get five marketing jobs, one prompt each, and the real output we got on 12 November 2026. We run Weir, the read-only server used in every example. It cannot post.
Paste this into Claude or ChatGPT once Weir is connected:
Search X for original posts mentioning "GitHub Copilot" in the
last 7 days. Sort by top, 10 posts. Group them by theme and tell
me what people are asking for.What an MCP server is, in marketing words
An MCP server is a plug that lets your AI assistant fetch live data from one service. The Model Context Protocol is an open standard for connecting AI applications to outside systems, and Claude and ChatGPT both support it.
Claude calls them connectors.
Social media MCP for Claude: connect Weir in one click
Open the Add custom connector dialog with Weir filled in. It only fills the form. You check it, confirm, then sign in.
For ChatGPT, use the ChatGPT setup guide.
As of 12 November 2026, Weir gives 5,000 free credits a month, and $5 buys 25,000. An X post costs 1 credit, plus 1 per call.
Social listening MCP: what people say about a brand this week
Ask for original posts that mention the brand, and say whether you want newest or most popular.
Search X for original posts in English that mention "GitHub
Copilot" since 5 November 2026. Show 8 posts. For each, say the
author type and the topic. Then give me the three themes.We ran this through search_x on 12 November 2026. It returned 8 posts for 9 credits. They were from the last 90 minutes, and none had more than 9 likes.
That is the catch with newest-first: it shows now, not this week. The 8 posts were a new-app announcement, an API change, a tier list, an outage comment, a hiring post and adverts.
For a weekly mood read, ask for top posts, as in job 4. Cost: 9 credits.
Social media MCP for competitors: their best posts this month
Search the account's own posts with sort: top and replies removed to get a shortlist.
Show @github's original posts from the last 30 days, sorted by
top, 6 posts. Make a table with date, likes, views and what the
post is about. Then tell me what the best ones have in common.We ran it on 12 November 2026 with a start date of 13 October. It returned 6 posts for 7 credits. Sorted by likes in code:
Four of the six posts link to an engineering or security write-up. The other two are event videos. The top post states a number in its first line, and it got 85,359 views.
The order that came back was not strictly by likes, because X's top is its own ranking. Ask your assistant to sort the table. Cost: 7 credits.
Best MCP for marketing research: questions people ask, to turn into content
Search a category, keep the questions, and turn each theme into a post.
Search X for original posts that contain "code review" and
"how do you", since 1 November 2026, English, 8 posts. List the
questions people ask. Group them. Suggest one content idea each.We ran it on 12 November 2026. A first try with extra operators returned nothing and cost 1 credit. The simpler query returned 8 posts for 9 credits.
Four were real questions. Two asked how to give review feedback kindly. Two asked how to keep up with AI-written code.
The rest was noise, such as a plumbing advert. Tell the assistant to drop it. Cost: 9 credits.
Who is talking about a topic, by follower count
Search the topic, ask for top posts, and sort the authors by followers. We name only brands here.
Search X for "GitHub Copilot", original posts only, sort top,
since 5 November 2026, 10 posts. Make a table of author
followers, likes and topic. Sort by followers.We ran it on 12 November 2026. It returned 10 posts for 11 credits. Here are four of them, sorted by followers in code:
| Author | Followers | Likes | Post about |
|---|---|---|---|
| @github | 2,737,993 | 301 | A new model in Copilot |
| A developer-tools curator | 100,930 | 25 | A directory of agent skills |
| @luganodes | 12,622 | 4 | Its own MCP server launch |
| A founder with a poll | 4,221 | 29 | Which AI coding tool to pay for |
Followers and attention do not match. The founder's poll had the fewest followers of the four, yet it drew 41 replies, the most of any non-brand post. Cost: 11 credits.
Replies under a launch post, grouped by theme
Replies show what the audience did not understand. Give the assistant a post URL and ask for themes.
Get the replies to this post: https://x.com/github/status/
2111880011485516083. Sort by likes, 15 replies. Group them by
theme and count each group. Quote nothing from private accounts.We ran it on 12 November 2026. It returned 15 replies for 16 credits. The post said GitHub's Security Lab used an open-source agent to find 24 Android bugs. The replies grouped like this:
The other 3 were two jokes and a link-only reply.
The five questions are a ready-made FAQ. Cost: 16 credits.
Posting and scheduling need a different kind of server
Weir is read-only. It cannot post, reply, like or schedule.
For posting you need a server from a social media scheduling and publishing tool, or an official platform API with write access, exposed through MCP. We have not tested one, so we recommend none. Check the docs for a tool that creates a post, and read the permission screen first.
Pair the two: read-only to decide what to write, a publishing tool to post it. Weir also has Reddit tools, on the Reddit MCP server page.
Run it daily: a monitor that sends new posts to a webhook
A monitor runs the same search on a schedule and reports only what is new. As of 12 November 2026, our monitors docs list manual, hourly, daily and weekly schedules. Hourly is paid only, and the free tier gets one monitor.
Each run costs a 1-credit minimum plus the new results it stores. A search monitor reads one page of about 20 posts, so a daily run costs about 21 credits at most. Thirty days is about 630 credits, inside the free 5,000.
Create one with POST /v1/monitors, using "kind": "search", your query and "schedule": "daily", as in the docs. We did not create a monitor on our research account.
Point a webhook at it and each run with new posts sends one delivery of up to 25 items. A Slack, Discord or Telegram webhook URL works too. More in our guide to an AI social listening agent and the social listening API page.
When not to use Weir
- You need to post or schedule. Weir is read-only, so pick a publishing tool.
- You need Instagram, TikTok or LinkedIn. Weir reads X (Twitter) and Reddit only.
- You need a contract with a platform. Weir is not affiliated with X or licensed by it, so access can change.
- You track private people. Weir reads public data only, and we do not help profile individuals.
Check each platform's terms before you collect data. This is not legal advice.
FAQ
What is MCP for marketing?
It is a way to connect Claude or ChatGPT to live marketing data through an MCP server. You ask in plain words and the assistant fetches and summarises the data.
What is the best MCP for marketing?
It depends on the job, so we do not rank them. Pick one server per job: ads data for ad reports, a publishing tool for posting, a read-only server like Weir for listening and research on X.
Is there a social media scraping MCP I can use in Claude?
Weir is one for public X posts and profiles. You connect it in Claude with a one-click link, and it returns each post's text, counts and author. It reads public data only.
What can a social listening MCP not do?
It cannot see private accounts, deleted posts or who liked a post. It also returns what a search finds, which is not every post. Treat themes as a sample.
Sources
Checked 12 November 2026:
- Model Context Protocol, what MCP is: https://modelcontextprotocol.io
- Weir monitors and webhooks docs: https://docs.weirapp.io/monitors
- Weir Claude setup, one-click link: https://weirapp.io/integrations/claude
- Our @github and brand searches through Weir's MCP tools, 12 November 2026
- Sociality, "What is MCP for marketing?" (updated 7 October 2026): https://sociality.io/blog/mcp-for-marketing/
- Vector, "We built an MCP for marketing" (6 May 2026): https://www.vector.co/blog/we-built-an-mcp-for-marketing-heres-what-it-actually-does
- LimeLight Marketing, "Claude MCP for Marketers" (20 August 2026): https://limelightmarketing.com/blogs/claude-mcp-for-marketers/