A Claude Code social media plugin is useful when research should become a file in your project: a product brief, a launch issue list or a CSV of reviewed sources. The valuable step is connecting retrieved evidence to the work you are already doing, with the search method still visible. A paragraph in chat is only the start.
Consider a developer preparing a microphone-comparison feature. Reddit discussions may reveal setup problems and YouTube comments may reveal unanswered compatibility questions. The agent can help gather a bounded sample, but the deliverable should show exactly which sources informed a proposed requirement. This playbook takes that question through a small project-based workflow.
Separate the connection from the workflow instructions
Anthropic’s plugin guide describes packages that combine skills, connectors and other components. A plugin can teach a repeated method; an MCP connection supplies external tool access. Installing a package does not automatically authorize every source, every search budget or every local file operation.
The Superlurk Claude Code plugin includes a search skill and the commands /superlurk:search, /superlurk:results and /superlurk:balance. Use the current Claude Code integration instructions for installation and authentication. Do not assume an old command from a copied blog or a marketplace listing applies to the version and access you have.
Prepare a small research workspace
- Write the question. State the decision the research should inform: which microphone setup questions need an answer before a user buys? Keep it to one product task.
- Choose the output location. Use a dedicated folder such as research/microphones. Name the proposed brief and evidence files before allowing the agent to write them.
- Define selection rules. Choose the networks, relevant product generation, language and exclusion rules. Decide what counts as first-hand experience and what needs corroboration.
- Set a stop condition. Ask for a small number of reviewed examples or a maximum approved credit budget, whichever is reached first. Leave room to inspect a useful thread.
Do not place secrets, support exports or customer identifiers in the research prompt merely because the agent is running inside your repository. A public-search question usually needs the category and the problem, not a production database connection. Ask the agent to read only the local files relevant to the brief.
Start with free discovery and a priced plan
Use /superlurk:balance to understand the available balance and connection cap, and /superlurk:results to see whether earlier work already contains useful evidence. Ask the agent to discover the appropriate searches and show their quotes before it spends credits. Reading your own saved Superlurk results is free; a fresh search or refresh can cost credits.
Research public questions about USB microphone setup on Reddit and YouTube. Reuse relevant saved results first. Show a small search plan and the exact quotes before spending. After approval, inspect the useful sources and write research/microphones/brief.md plus an evidence CSV. Label assumptions and unavailable context. Do not change application code or publish anything.
This prompt scopes the research and file outputs. Claude Code permissions remain a separate control. Anthropic’s security documentation explains how permission modes and workspace controls affect actions. Review proposed operations in the mode you use; do not depend on a screenshot of somebody else’s approval dialog.
Turn source retrieval into a decision record
Ask for an evidence row for every observation: network, source URL, publication date when available, retrieval date, query, observed text or behavior, interpretation and review status. When a reply changes the meaning of a post, keep that connection. A quote about an older product version should not silently become a requirement for the current one.
The brief should have a concise answer, the strongest supporting observations, counterexamples and remaining questions. A useful recommendation might be “test whether users can identify the correct input device” rather than “Reddit hates this microphone.” The first can lead to a reproducible product test; the second overstates a selected sample.
- Record why a source was included and which product/version it describes.
- Mark first-hand reports separately from promotion, hearsay and quotations of another post.
- Keep a requirement or hypothesis separate from a directly observed fact.
- Do not turn engagement counts into market share or counts of distinct affected customers.
Export without losing provenance
The plugin can help write CSV evidence in the current project; the web application also offers CSV and JSON exports from saved Results. Neither format guarantees that the assistant has included every item it could have read. Record which saved result and pages were used, and preserve the original extract alongside any cleaned worksheet.
Do not commit a research export automatically. It may contain usernames, public text or information unrelated to the codebase. Review its necessity and your sharing rules, then decide what belongs in version control. The structured export guide gives a field-level handoff method.
Review the brief before implementing a change
Open a few source links, including the source supporting the most consequential recommendation and a counterexample. Check that every quote is traceable, every inference is labeled and every missing field remains missing. If a source can no longer be viewed, preserve that limitation rather than asking the agent to fill it from memory.
Once the evidence is reviewed, make implementation a separate explicit task with its own acceptance criteria. You can carry the brief into an issue or design document without granting public posts authority over the repository. Read social MCP security for handling instructions embedded in retrieved material.
Frequently asked questions
Does Claude Code need a plugin to use MCP?
No. A direct MCP connection can expose tools. A plugin can bundle that connection with task instructions and commands, which may make a repeated research workflow easier.
Can the Superlurk plugin publish social posts?
No. Its social tools retrieve public data. A local brief or CSV file is an artifact in your project, not a published social post.
Do slash commands use different credits?
No. Superlurk searches use the same account, balance and prices as the web application. Review the quote before running a search.
Is every Claude plugin feature available in every Claude surface?
No. Anthropic documents differences between chat, Cowork and Claude Code. Check the components your workflow needs and the current client instructions.
Sources
How this guide was written. Reviewed on 11 November 2026 against Superlurk's accepted search catalog, integration instructions and the primary references below. The workflows and example budgets are editorial examples, not customer results or a new API contract. Availability, source visibility and client setup can change; use the linked integration page for current connection instructions.