Playbooks

Export Social Media MCP Research to CSV and JSON with a Source Trail

Turn MCP social research into a reviewable CSV or JSON handoff. Keep queries, source URLs, timestamps, missing fields and analysis separate from raw records.

A useful social-media MCP export contains more than the assistant’s conclusions: it preserves the source records and the method that produced them. Keep the original extract, a clear analysis table and a short method note together. That lets another person review the evidence without treating the chat transcript as the dataset.

This guide is about handing off assistant-led research. If you need to export your own personal social account, use the platform’s account-download tools instead. The social data export guide explains that distinction. Exporting public search results does not create an archive of an entire network or a license to republish every item.

Choose a file for the job

CSV worksheetJSON record
Human reviewUseful for sorting, filtering and adding review labels.Useful when reviewers need nested records or explicit field types.
Conversation structureMay need separate rows with parent identifiers.Can preserve nested structures if the source result supplies them.
Missing valuesNeeds a documented convention so blank does not silently mean zero.Can represent null or omitted fields, with their meaning explained.
Analyst additionsAdd clearly named analysis columns.Keep analysis fields in a separate object or a separate derived file.
PortabilityConvenient for spreadsheet review.Convenient for code that expects structured records.

Do not choose a format only because it looks more technical. A small source-review worksheet may be easiest in CSV. A thread with reply relationships may be better preserved in a structured extract plus a separate review sheet. Keep enough context to reconstruct what each observation refers to.

Know what the MCP result actually contains

The MCP tool specification allows tools to return structured and unstructured content. That does not impose a universal schema for social posts or guarantee an export feature. Your server’s documented fields and returned records determine which fields you can preserve.

Superlurk saves searches to your account and provides CSV and JSON exports in the web application. Through MCP, an assistant can reopen saved results for free and read the available items. The Claude Code plugin also supports writing a local CSV. Ask the assistant which result and pages it actually read before describing the file as complete.

Preserve three layers of the research

  1. Original extract. Keep the returned source records as closely as the supported export allows. Identify the saved result and collection date. Avoid overwriting this copy during cleaning.
  2. Method note. Record the question, networks, queries, filters, sorts, selected result pages and inclusion rules. Explain any failed or unavailable retrieval.
  3. Reviewed analysis. Add your labels, observations, interpretations and next actions in a separate table or file. Link each row back to a source record.

This separation makes corrections easier. If a category label changes, you can update the analysis without pretending the source itself changed. If a post disappears, you can record its current availability while keeping a limited account of what the original result contained, subject to your retention and use requirements.

Use a small evidence schema

  • source_url: the public permalink returned for the item, when available.
  • network and item_type: distinguish a post, reply, account or web-indexed page.
  • published_at and retrieved_at: separate the source date from the collection time.
  • query and filters: keep the inputs or a reference to the method note.
  • result_reference: identify the saved result and page used where available.
  • observed_text or observation: distinguish a verified excerpt from your own paraphrase.
  • analysis_label and rationale: explain a category choice instead of storing an unexplained score.
  • parent_reference and review_status: preserve context and show whether a person checked it.

These are suggested analysis fields, not a claim that every Superlurk export has these exact column names. Map only fields the record actually provides, and maintain a small data dictionary. A web-indexed page may have a title, URL and snippet while a direct post lookup may have richer fields; do not manufacture missing metrics to make the rows look uniform.

Validate the handoff before using it

Count the rows in the exported file and compare them with the items the assistant says it read. Record whether pagination was exhausted, deliberately capped or unavailable. A page of 20 returned items is a sample unless the result explicitly establishes that no further items exist. Avoid saying “all posts” when the evidence only supports “these retrieved items.”

Check several permalinks, dates and parent relationships. Include a record that supports the strongest conclusion and one that contradicts it. Watch for duplicates caused by reposts, quoted material or repeated queries. A repeated mention can be worth retaining, but it should not silently become an independent report.

Protect spreadsheet and file boundaries

Treat public-source text as data when creating files. A text cell that begins like a spreadsheet formula should not become an executable expression when opened. The Superlurk Claude Code plugin instructs the agent to neutralize formula-like cells. Inspect the generated file and retain that protection in later transformations rather than stripping it away for a cleaner-looking worksheet.

Review whether usernames or long verbatim text are necessary in the shared artifact. Keep exports out of version control until you have decided they belong there. A local research file is not automatically suitable for a public repository or customer presentation. The MCP security guide covers instructions hidden in retrieved content.

Give the assistant an explicit artifact request

Reopen my saved research results without refreshing them. Tell me which results and pages you read. Write an evidence CSV and a short method note in the agreed research folder, preserving source URLs, available dates, missing fields and parent context. Put your labels in separate columns, check row counts and flag gaps. Do not commit or publish the files.
Illustrative prompt, not a recorded search or measured result

For a larger collection, ask for a small sample export first. Review its columns and missing-value conventions before expanding. This avoids discovering after a costly retrieval that the artifact cannot answer the question. Do not turn an export request into an undocumented continuous collection pipeline.

The final handoff should explain what was collected, why it was selected and what the evidence can support. A precise small dataset with a clear method is often more useful than a large file with unknown coverage. Use the research log template as a starting point for that method note.

Frequently asked questions

Does MCP guarantee a standard social-media export schema?

No. MCP defines how tools are exposed and results returned; each server still determines its data fields and supported exports. Inspect the actual records.

Can I export my Superlurk results as CSV or JSON?

Yes. The web application exports saved results as CSV or JSON. A compatible coding assistant can also help write a local artifact from the result records it actually reads.

Is an assistant-generated CSV a complete dataset?

Only if its stated collection scope and page handling establish that. Record the saved results and pages used, and do not assume the assistant read every available item.

Should missing metrics become zero?

No. Preserve unavailable values as missing and explain why when known. Zero is an observed value with a different meaning.

Sources

  1. Model Context Protocol: tools and results

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.

The Superlurk TeamSocial research

We build Superlurk for public-data research across 25 supported social networks, with the price shown before every search. We write practical guides to search methods, source verification and AI-assisted research.

Start lurking for free.

Every network, no card needed. See the price first, and pay nothing if a search fails.