To search social media with an AI assistant, connect a compatible research tool through MCP, choose supported searches, review their credit quotes and keep the source links with the answer. Superlurk combines network-data searches and public pages found on the web; what comes back depends on the selected action. This guide explains the workflow, with separate setup instructions for each client and practical limits on cost and coverage.
networks your assistant can search
what most searches cost, quoted before they run
what a failed search costs
free trial credits, and the trial can connect too
What MCP is, in one paragraph
MCP, the Model Context Protocol, is an open-source standard for connecting AI applications to external systems. Its documentation compares it to a USB-C port: one standard way for assistants such as Claude and ChatGPT, and tools such as VS Code and Cursor, to plug into data sources and tools. A service publishes one MCP server, and any client that supports MCP can connect to it. For social search, that means one server gives several assistants the same searches, the same prices and the same saved results.
What your assistant can do once connected
Superlurk's MCP server offers a handful of simple tools rather than one tool per search, and the assistant chains them for you:
- Find the right search among the 492 ready-made searches on TikTok, Instagram, YouTube, X, Reddit, Threads, LinkedIn, Telegram, Lemon8, Bluesky and 15 more sites.
- Get the exact price of that search with your inputs, before anything runs.
- Run the search from that price quote. This is the only tool that spends credits.
- Check a search that is still running and return the first page of results.
- List and reopen your saved results, for free.
- Check your balance, your plan and this connection's daily cap, if you set one.
The searches are the same ones as in the web app, so the same limits apply: LinkedIn actions include link-based lookups, company and person posts, post comments and job searches, Instagram keyword discovery searches accounts, hashtags and places rather than every caption, and only public content can be searched.
How to connect it, client by client
Every client uses the same server address and the same sign-in. You sign in to your Superlurk account in the browser with OAuth, so there is no API key to copy or paste, and every plan can connect, including the free trial. The server URL and the current, exact steps are on the integrations page. Here is what to expect in each client.
Claude (web, desktop and mobile)
In Claude on the web or desktop, open Customize, then Connectors, and add a custom connector named Superlurk with the server URL. Choose Connect, sign in to Superlurk and approve, then switch Superlurk on in a chat's tools menu and ask. Once added, it is available in the Claude mobile apps too. Anthropic's help centre says custom connectors are available across Claude's plans, with one custom connector on the free tier, and that on Team and Enterprise an owner adds the connector for the organisation before members connect. Full steps: Superlurk for Claude.
ChatGPT
In ChatGPT on the web, open the plugins page, choose to add a custom MCP server, name it Superlurk, paste the server URL and choose OAuth for authentication. Create and install it, then sign in to Superlurk and approve. In a chat, type @Superlurk and ask; prices and results appear as cards in the conversation. OpenAI notes that workspace permissions and security settings apply, so in a company workspace you may need an admin's help. Full steps: Superlurk for ChatGPT.
Claude Code
Use the Claude Code integration guide for the current remote-server command, plugin setup and browser sign-in. Review Claude Code’s tool permissions and ask it to show the quote before execution. Approval behavior depends on your client settings; set a daily connection cap when you need a spending limit.
Cursor
Use the Cursor integration guide for its current MCP configuration, Customize controls and OAuth sign-in. Keep credentials out of project files, inspect the available actions and begin with a small priced research question.
VS Code
Follow the VS Code integration guide to choose the correct MCP configuration format and sign in. Its guided server setup is preferable to copying a configuration for a different location or release. Review tool permissions before requesting a search.
What happens on your first search
- You ask in plain words. For example: “What are people on Reddit saying about standing desks this week? Show me the price first.”
- The assistant finds the search. It picks the network and the ready-made search that fit, such as Reddit post search, and fills in the inputs.
- You see the price. The assistant gets an exact quote, typically 1 credit, and tells you before anything runs.
- Review the quote and execution permission. Approve the priced action through your client’s supported workflow. Client settings determine when a permission prompt appears; a failed search is free.
- The results are saved. Every result goes to your private Results in Superlurk, where you can reopen it for free or export it as CSV or JSON.
Example prompts
- “What are people on Reddit saying about standing desks this week? Show me the price first.”
- “Find the most-viewed TikToks about matcha latte recipes and tell me what they have in common.”
- “Search TikTok and YouTube for reviews of compact espresso machines, one search each, and list the problems people mention.”
- “Show me today's trending searches on TikTok and Reddit and suggest three video ideas for a home coffee channel.”
- “Compare how Threads and X are reacting to the new phone launch. Stay under 10 credits.”
- “Search X for posts about our launch since Monday and save them to launch-posts.csv.” This one suits Claude Code, which can write the file in your project.
- “Open my saved Instagram hashtag results from yesterday and turn them into a table.”
- “How many Superlurk credits do I have left?”
Price first, and an optional daily cap
An assistant that can spend credits needs guard rails. Superlurk's are the same as in the web app, plus one more:
- The price comes before the spend. A search can only run from a price quote the assistant fetched first, so it always knows the cost and can tell you before asking to run.
- The spending step is identified. Running a search is flagged as a spending action. Review your client's tool-approval settings and ask it to show the quote before you approve. Do not assume a prompt alone enforces a hard spending cap.
- Same prices as the app. Searches cost 1, 5 or 25 credits depending on the action. A failed search is free, and reopening a saved result is free.
- An optional daily credit cap per connection. When you sign in from an assistant you can set a daily cap for that connection, then add, change or remove it later under Settings › AI connections. A search above what is left of the cap is refused. There is no cap unless you set one.
- Purchases stay on the website. Nothing is sold inside the assistant. If your balance runs low, you top up in Superlurk.
For more on how credits work, see credit pricing for social search.
Privacy: what the assistant can and cannot see
- Only your own account. Through the connection, the assistant sees the search catalogue, your prices, your balance and your own saved results. It never sees other people's results, your payment details or your password.
- No keys to leak. Sign-in uses OAuth in your browser, so no API key sits in a config file or a chat.
- Revoke at any time. Remove the connector in the assistant and revoke it under Settings › AI connections in Superlurk; revoking stops it straight away.
- Public content only. The assistant can only search what each network makes publicly available, and results stay private to your account.
- Your assistant's own terms apply. Results you bring into a chat are handled under that assistant's settings and policies, like anything else you share in a conversation.
- Research topics, not people. Use it for brands, products and public conversations, never to track private individuals.
Tips for better answers
- Name the network when you know it, or ask the assistant to suggest the two or three most useful ones for your question. The guide to searching several networks at once explains which network answers which kind of question.
- Give a time window, such as “this week” or “since Monday”, so it picks a search with a date filter or latest ordering where one exists.
- Set a budget in the prompt, such as “stay under 10 credits”, even if you also have a daily cap.
- Ask for links to every post it quotes, so you can check them yourself.
- Reuse saved results. Ask it to open yesterday's result instead of running the search again; reopening is free.
- Export when you need a file. Selected results export as CSV or JSON from Superlurk, and the guide to exporting social media data shows how to work with them.
How answer engines use public social data
An answer engine can retrieve accessible web pages, select passages and use them to support a response. A public social post may be discoverable through an indexed page, a cited article or a connected retrieval tool, but those are different routes with different gaps. A useful answer should name the route and link to the original item where possible. Google's guidance for AI features in Search says supporting pages need to be indexed and eligible for a search snippet. It does not promise that every public post or useful page will appear in an answer.
For a research task, separate the engine's initial answer from the evidence you verify. Open the source, check the author and date, and confirm that the cited passage actually supports the specific claim. An old indexed excerpt may outlive an edited or deleted post. Superlurk's MCP connection provides its supported searches to compatible clients; it is not a claim that Perplexity or every answer engine has a hosted Superlurk connection. Neither web indexing nor MCP grants private-group access, private account analytics or a complete view of a network. Label missing coverage before turning a small public sample into a broader conclusion.
Use a discovery, quote and evidence loop
Ask the assistant to discover the actual search before constructing inputs. A community-post lookup, keyword search and account inspection have different requirements. Where a result offers a supported follow-up, review its filled inputs and obtain a separate quote rather than assuming that opening the next conversation step is free.
For a known list, eligible bulk lookup can accept up to 50 accounts, links or IDs. Review included and skipped items and the total before approving. Each item is its own lookup, so a failed item does not mean that successful ones were free. See bulk lookup preparation.
Keep a short evidence ledger with the question, query, publication and retrieval times, source URLs and limitations. Ask the assistant to identify which claims it verified and which are interpretations. A citation should support the exact sentence beside it, not just lead to a page on the same broad topic.
For the protocol and action model, use the social-media MCP guide. For sign-in, caps or empty-result problems, use the troubleshooting checklist.
Monitoring versus AI-assisted research
A monitor repeats a defined search on a schedule and saves each run's results to your Results. AI-assisted research is a conversation in which an assistant can help select searches, inspect evidence and draft an interpretation. One does not replace the other: an assistant session is not continuous monitoring, and a scheduled search does not establish what a new post means.
Use a monitor for a stable recurring question with a named reviewer. Use an assistant when the next research step depends on what you discover, while keeping prices and source links visible. If you need press coverage, licensed news archives or an operational crisis response, assess those requirements separately. Do not describe public social-search tools as a complete media-monitoring service.
For a fuller distinction, see social listening versus monitoring. A useful setup says which source and query are repeated, when the reviewer checks them, and when an assistant may propose a separately priced follow-up.
Use the right connection for a coding project
Choose the setup page for Codex, Cursor, VS Code or Claude Code. For Claude Code, the social research plugin workflow explains the relationship between the plugin, its skill and the MCP server. Keep fetched social posts as evidence, and use the security checklist before allowing a research task to write files or invoke other tools.
Frequently asked questions
Can ChatGPT search TikTok or Reddit?
A connected research tool can give ChatGPT supported public-data actions on TikTok, Reddit and other networks. Choose the Superlurk app or connection in your client, review the available actions and follow the current integration guide. Some networks return network data; others return public pages found through web search.
What is a social media MCP server?
An MCP server that gives AI assistants tools for searching social networks. Superlurk's server lets Claude, ChatGPT, Claude Code, Cursor, VS Code and other MCP clients find the right search, get its price, run it and reopen saved results.
How do I search TikTok from Claude?
Add Superlurk as a custom connector in Claude under Customize, then Connectors, sign in to Superlurk and approve, switch it on in a chat and ask, for example: “Find TikTok videos about matcha latte recipes and show me the price first.”
Will the assistant spend my credits without asking?
Superlurk quotes searches before execution. Client approval behavior depends on the assistant and your settings, so do not rely on a conversational request as your only spending control. Review each quote and set a per-connection daily credit cap where needed.
Do I need an API key?
No. The first time you connect, you sign in to your Superlurk account in the browser with OAuth, so there is nothing to copy, paste or leak.
Does it cost extra?
Superlurk uses the same credits and prices in the web app and MCP. Most searches cost 1 credit; web-sourced searches cost 5 and some deeper actions cost more. Failed searches are free. Your assistant may have its own subscription requirements.
Sources
How this guide was written. Substantively reviewed by the Superlurk team on 8 November 2026 against the accepted search catalog and the primary references below. Examples illustrate a research method, not observed customer outcomes. Availability and quoted prices can change.