AI assistants

Tagaroo in ChatGPT and Claude

Ask your assistant for a kappa, a reliability sample size or a COREQ check and it runs Tagaroo's tested implementations instead of guessing. Paste an interview and it codes the transcript, then shows the result in an annotator you can edit and open in Tagaroo. No account, nothing stored.

Connect

Claude

In claude.ai or Claude Desktop open Settings → Connectors → Add custom connector and enter https://api.tagaroo.ai/mcp. No sign-in is required.

Claude Code, as a plugin with skills: /plugin marketplace add Kikolo3000/tagaroo-claude-plugins then /plugin install tagaroo@tagaroo. Or the server alone: claude mcp add --transport http tagaroo https://api.tagaroo.ai/mcp

ChatGPT

Find Tagaroo in ChatGPT's app directory and connect it, or mention Tagaroo in a chat. The ChatGPT version covers research methods and annotation; the clinical scoring and de-identification tools are Claude-only.

Tools

Annotation

  • get_coding_scheme

    A ready-to-use scheme: the demo codes or a Scale Library instrument's categories and ladders, with coding instructions for the assistant.

  • annotate_transcript

    Places the assistant's codes on exact spans and opens the interactive annotator, where you edit codes and severities and open the work in Tagaroo.

Reliability and study design

  • compute_agreement

    Cohen's, weighted and Fleiss' kappa, Scott's pi, Gwet's AC1 and Krippendorff's alpha with confidence intervals.

  • compute_icc

    Intraclass correlation for continuous ratings, with the ICC form made explicit.

  • compute_span_agreement

    Agreement between coders who highlighted spans of text.

  • reliability_sample_size

    How many units to double-code for a target precision.

  • thematic_saturation

    Whether new interviews are still adding codes.

  • write_irr_report

    A reporting paragraph built from computed reliability results.

  • plan_qualitative_sample_size

    Information power and published saturation ranges for planning how many interviews to run.

  • compute_miti_scores

    MITI 4 summary scores (reflection-to-question ratio, % complex reflections, global scores) from behaviour counts.

Measurement

  • cronbach_alpha

    Cronbach's alpha with a Feldt interval, item analysis and alpha-if-deleted.

  • minimal_detectable_change

    SEM and minimal detectable change from test-retest data or published values.

  • screening_accuracy

    PPV, NPV and likelihood ratios for a screening test at a given prevalence.

  • convert_dice_iou

    Dice and IoU conversion, or full overlap figures from region counts.

  • word_error_rate

    WER, MER, WIL and CER between a reference and an ASR transcript.

  • convert_hamd_madrsClaude only

    HAM-D 17 and MADRS totals and improvements converted with published anchors.

Transcripts and reporting

  • clean_transcript

    Zoom, Teams, Otter or Rev exports into clean speaker turns.

  • check_reporting_checklist

    COREQ and SRQR items, and which ones a methods section already reports.

  • build_codebook

    A codebook in Markdown and as a Tagaroo taxonomy.

  • search_scale_library / get_scale

    Published coding scales and rating instruments with their categories, citation and licence note.

  • recommend_rating_scale

    A shortlist of instruments for a study design: construct, who rates, purpose and unit.

  • deidentify_transcriptClaude only

    Detects and redacts names, dates, contact details and identifiers.

  • score_rating_scaleClaude only

    PHQ-9, GAD-7, HAM-A, HAM-D, MADRS, YMRS and BPRS totals and bands.

Each tool mirrors one of the free Tagaroo materials, so you can check any result in the browser.

Annotating in chat

  1. Paste a de-identified transcript with speaker labels (for example I: and P:).
  2. Give your codebook, name a Scale Library instrument, or ask for the demo scheme.
  3. The assistant codes the transcript and shows it in the annotator, each code highlighted with its severity and rationale.
  4. Edit codes in place: change a severity, delete a code, or select text to add one.
  5. Press Open in Tagaroo to continue in the workbench, where you can keep annotating, compare coders and export to NVivo or ATLAS.ti.

Your data

  • Tool inputs are processed in memory to compute the result. Tagaroo stores none of them and never sees the rest of your conversation.
  • Open in Tagaroo carries the annotated transcript in the link's fragment, which your browser does not send to any server. It stays in your browser until you choose to save it to an account. If you are already signed in to Tagaroo, opening the link adds the transcript to your account directly.
  • Use de-identified transcripts only. Details are in the privacy policy.

Support

Questions, bugs or a tool you would like to see: write to support@tagaroo.ai. Include the assistant you used and, if you can, the prompt.

Frequently asked questions

Do I need a Tagaroo account?

No. Every tool works without signing in and is free to use. An account only comes into it when you open an annotated transcript in Tagaroo and decide to save it to a study.

Who does the annotating?

Your assistant. ChatGPT or Claude reads the transcript and codes it against your codebook or a library scale; Tagaroo finds the exact span for each quote and draws the editable annotator. Tagaroo's own annotation agent and curated coding manuals run only inside the Tagaroo app.

What happens to the data I send?

Tool inputs are processed in memory and returned to the assistant. Nothing is stored and the rest of your conversation never reaches us. Open in Tagaroo puts the annotated transcript in the link's fragment, which is read by the page in your browser and not sent to our servers.

Can I use it with patient transcripts?

Use de-identified transcripts only. In Claude, the deidentify_transcript tool can redact identifiers before you code. The ChatGPT version does not include the clinical scoring or de-identification tools.

Why are the statistics computed by a tool and not by the assistant?

Language models are unreliable at arithmetic over many rows, and reliability coefficients have several variants that are easy to mix up. The tools run the same tested implementations as Tagaroo's free calculators, so the number in your paper is reproducible.