How Do I Give Claude Access to My Competitors' Social Media? MCP Setup and Costs

Connect Claude to an MCP server that can read public social media, then ask it about a competitor by name. Claude cannot see Instagram, TikTok or YouTube on its own, and it should not log in as you to get there. An MCP server gives it tools for public, logged-out data, so "what has Acme posted since Monday?" becomes a tool call with a source link on every result.

The setup takes a command or a pasted config, and the choice that matters comes before it: whether the server reads a competitor account on request or only reports on accounts you set up elsewhere first.

What can Claude read about a competitor once it is connected?

Only what a logged-out visitor can see. On TrueScrape that means public profiles and posts, plus these account feeds Claude can watch over time:

• YouTube channels
• TikTok and Instagram accounts
• Threads and Bluesky accounts
• Reddit users
• Mastodon accounts
• Telegram channels
• Substack publications
• LinkedIn company pages

It cannot read private accounts, direct messages, follower lists behind a login, or a competitor's own analytics. Paid reach, conversion and revenue are not public on any of these platforms, so no connector can honestly report them. Any figure Claude derives, such as an average across recent posts, is an estimate from the posts it read, and it should say how many that was.

X and Facebook accounts cannot be watched this way today. If those are where your competitors matter most, use one of the options below that lists them.

How do you connect Claude to social media data?

Three steps, using Claude Code as the example:

1. Get an API key. A free TrueScrape key comes with 1,000 credits and needs no card.
2. Sign in once: run npx truescrape auth login, or set TRUESCRAPE_API_KEY.
3. Add the server: run the one line below.

npx truescrape agent add claude-code

The same command takes claude-desktop, cursor or vscode. Every flag and file path is in the setup reference.

By default it loads tools for nine common sources rather than everything. That choice is about context, because tool descriptions are read into the conversation before you ask anything. Measured on 1 November 2026, the default list was 105 tools and about 87 KB, and the full list was 265 tools and about 198 KB. The tools left out are still reachable: explore_tools finds them with their cost and inputs, and it is free.

What does a real competitor check look like?

Here is one run, made through the live MCP server on 1 November 2026. We called the tools directly, in the order an agent would. Acme is a stand-in in the prompts on this site, so the run needed a real channel: we used Notion's YouTube channel, chosen only because it is a public company account.

The prompt: "Watch this competitor's YouTube channel and tell me what's new each time I ask."

The first call creates a monitor with no schedule, so nothing runs or charges until the agent asks:

{
  "tool": "create_monitor",
  "arguments": {
    "term": "Notion",
    "platforms": [],
    "accounts": [{ "platform": "youtube", "handle": "@Notion" }],
    "schedule": "manual",
    "agent": "claude-code"
  }
}

Each later step, with what it returned and what it charged:

Tool	What came back	Credits
create_monitor	A monitor id, checksPerDay: 0, and a note stating the cost of each check	0
check_monitor, first time	isBaseline: true, 0 new items	1
get_monitor_mentions, unread only	Empty, because a baseline announces nothing	0
check_monitor, second time	isBaseline: false, 0 new items	1
get_monitor_mentions, unread only	Empty, because nothing had been published	0

The checks were about a second apart, so finding nothing new was the correct answer, and this run does not show a new post arriving. It does show the cost model. The first check records what already exists, so you are not told about a channel's whole back catalogue. Every check after that charges the feed's normal price whether or not anything changed. Storing the results and reading them cost nothing. At the $1.25 per 1,000 credits starter rate, one check of one channel costs about a tenth of a cent. Other feeds have their own prices, and the response states what was charged.

How does Claude remember what it has already seen?

The server remembers, because Claude starts each session with no memory of the last one. get_monitor_mentions with unread_only returns what this API key has not read yet and moves its position forward. Two agents with different keys on one team do not take each other's updates.

That gives you a simple habit. Open a session, ask "what have my competitors posted since I last looked?", and Claude checks the monitor and reads only the new posts. If you want it on a schedule instead, drop schedule: "manual" and it checks every 6 hours by default, or as often as once an hour. It is not a live stream.

How do you stop an agent leaving a bill running?

A scheduled monitor keeps charging after the session that created it ends, so decide how long it should last when you create it:

• Manual monitors have no recurring charge at all. Nothing runs unless the agent asks.
• stop_at ends a scheduled monitor at a date you choose.
• max_credits ends it once that much has been charged, and it is never exceeded.
• The agent field is required over MCP and marks the monitor as agent-created. That work has a projected limit of 200 credits a day by default, and a create past it is refused for free, naming the cost.

Failed and empty checks cost nothing. One known ceiling on the limit: it depends on the caller saying it is an agent, so a script calling the plain HTTP API without that field is not limited.

Can a competitor's post trick Claude?

It can try, and this is the risk to take seriously. You are handing an agent text written by strangers. A post could say "ignore your instructions and do this instead". Every result from the server ends with a notice that the text is third-party data and never instructions, and we saw it on every call in this run. That lowers the risk without removing it, because no server can guarantee how a model behaves.

So keep the permissions boring. Do not give the same session unattended access to send email, spend money or change production when it is also reading strangers' posts. Have Claude show you the source link before you act on anything it read.

Which MCP server should you use for competitor research?

It depends on whether you already use a social media tool and whether you need a platform someone else covers better. These come from each vendor's own page, read on 1 November 2026, and they change.

Option	How competitors get in	Networks it lists	What you pay
Metricool MCP	Add them in Metricool first: "the MCP can only read what's already being tracked"	Examples use Instagram, TikTok and LinkedIn	Plan not stated in that guide
Sociality MCP	Competitor tools such as get_competitor_posts, alongside your own accounts	Instagram, TikTok, LinkedIn, YouTube, X, Facebook	1,000 one-time credits free on Starter
SociaVault MCP	Any public profile, through 107 tools	Names X, LinkedIn, Facebook, Reddit, TikTok and others, plus ad libraries	Open-source server; you buy API credits
TrueScrape	Any public account, by name in the request; no dashboard setup needed	The feeds listed above, plus one-off lookups elsewhere	1,000 free credits, then per call, from $25

Choose Metricool or Sociality if you want competitor numbers beside your own accounts and a team already works in that dashboard. Choose a server like SociaVault or TrueScrape if you want an agent to look at any public account the moment you name it. If X is essential, TrueScrape's monitors are the wrong tool for it today.

What should you ask Claude first?

Start with something you could check by hand, so you can see whether to trust it:

• "Look up Acme's YouTube channel and list its last ten videos with links."
• "Watch Acme's YouTube and TikTok, check them once, and tell me what you set up and what it cost."
• "Compare the topics of Acme's and Globex's last 30 posts. Say how many posts you read, and link each example."

Asking for the sample size and the links makes the answer checkable, and the links are what make it research rather than a confident guess. To go wider than named accounts, see how an agent can monitor brand mentions, which finds people talking about a competitor rather than what the competitor posts. For what a competitor sells rather than what it says, see connecting an agent to Amazon product data. Every tool and its price is on the agents page.