channel_breakdown
ChatGPTHow a project's responses split across collection channels (link/website/email/qr-code/social-media/panel), with counts and average completion progress.
check_design_snapshot
ChatGPTVerifies an uploaded design snapshot: the object must exist and be non-empty, self-contained HTML within the size limit. Returns { ok, previewUrl, sizeBytes, needsConfirmation, message }. needsConfirmation: true means the user should look at previewUrl and confirm it is the intended design. A missing, empty, oversized or non-HTML file returns an error; re-upload and check again. After a successful check: propose_usability_test(uploadId) for a new test, or relaunch_usability_test(projectId, uploadId) to re-test an existing project.
company_data
ChatGPTThe caller's own company profile: company name, a summary of what they do, their category, and their products/services (with features, benefits and unique selling points). Use this to GROUND ideation and strategy (e.g. "new project ideas", positioning, what to ask next) in what the company actually offers. No arguments — it is always the caller's organization.
contact_detail
ChatGPTOne contact (respondent): their name, email, phone, and recent responses across all projects (insight id + share id, project, summary, signal, date). Resolve contact ids via contact_list.
contact_list
ChatGPTList the organization's contacts (respondents) — optionally limited to those who responded to specific projects. Each contact shows their name, email, total responses, last response date, and which projects they responded to. Use projectIds to find who responded to given projects, or search to find a person by name/email. Pass a returned contact id to contact_detail or responses_search (contactId).
cross_project_comparison
ChatGPTCompare feedback across projects. Pass explicit projectIds (2+) to compare those. Pass a single projectId (or none) and the tool finds the most comparable projects by theme similarity. Returns each project headline plus shared and divergent themes.
export_responses
ChatGPTExport the RAW individual survey responses for one or more projects as a downloadable xlsx — one row per respondent, one column per question (scores, ratings, multiple choice, open text) plus email and completion, matching the in-app "export responses" download. Data is fetched server-side (you provide no rows). Each project becomes its own worksheet and carries its own optional filters (date range, sentiment, channel, quality, completion, read state, keyword); repeat a project id with different filters for separate sheets. Use this when the user wants all the underlying answers in a file. Returns a signed download URL once the file is ready; for a large export it returns { downloadId, status: "processing" } — then call get_export with that downloadId to get the link. The signed URL is SHORT-LIVED — read the exact expiry from the expiresAt field, tell the user to download promptly, and re-run get_export for a fresh link if it has expired. Do NOT invent a URL or an expiry. Response export is a PAID feature: if the organization's plan does not include it the tool returns an error saying so — relay that reason to the user verbatim, do not retry.
flag_result
ChatGPTFlag a synthesized usability issue as inaccurate or hallucinated to help tune the models. Pass the testId and the issue id from get_test_results; reason is optional.
generate_chart
ChatGPTRender a chart to a hosted image; returns a URL to embed in markdown as . Types: bar (grouped vertical), horizontal_bar (best for long category names or ranked lists), stacked_bar (part-to-whole across categories), line / area (trend over an ordered axis), pie / doughnut (share-of-total of ONE series).
get_export
ChatGPTFetch the status and download link for an export created by export_responses, by its downloadId. While the file is still being prepared this returns status "processing"; once ready it returns a fresh signed download URL that is SHORT-LIVED — its exact expiry is in the expiresAt field. Tell the user that real expiry, and call get_export again for a fresh link if it has lapsed.
get_question
ChatGPTFull detail of ONE question: its type, text, 1-based position (order) among the project's questions, the total question count, options, contentUrl, AI follow-up depth, and — for task questions — whether the success/feedback follow-ups are enabled. Use it to inspect a question's current position, options, and settings. Resolve question ids via project_overview's questions list.
get_test_results
ChatGPTReturns the usability results for a test. resultsMarkdown is a display-ready summary of every issue (summary, severity, a paste-ready fix prompt for a coding or design agent such as v0, Cursor or Lovable, and a conversation link) plus the instant design scan. Structured fields: prioritizedIssues[].{title, severity, detail, fixPrompt, conversationLink}, quickScan, phase and message. Results are final only when phase is "done". For a "partial" or "failed" run it returns the partial findings and a message describing what happened. wait_for_results returns the same data once a test finishes.
get_test_status
ChatGPTReturns a one-off progress snapshot for a running usability test: phase, an approximate percent complete and, for a failed run, failureReason ("partial" means the run was limited). Runs normally take several minutes. wait_for_results is the usual way to follow a test to completion.
insight_video_metadata
ChatGPTVideo-derived UX signals extracted from ONE usability-test session recording: findings (claims about the product, with polarity/severity and supporting quotes), frictions (taxonomy-classified problems with outcome, recovery action, channel and timings), task performance (per-task outcome and operational/perceived/efficient success), emotional moments (fear/anger/frustration/delight with triggers), and mental-model gaps (belief vs actual behaviour). Look it up by insightId (from responses_search / recent_responses / response_detail). Only usability-test responses whose recording was processed by video-metadata extraction have this; other responses return empty facets.
launch_usability_test
ChatGPTStarts the AI usability test for a proposal from propose_usability_test or revise_usability_test. Intended to run only after the user has reviewed the plan and explicitly approved launching it, because it spends AI-tester runs and creates a new project. Requires the latest proposalId: each revise replaces earlier ids, and an outdated id fails with PROPOSAL_NOT_FOUND. Optional last-minute edits: add or remove tasks or questions, swap personas. Returns the testId, the final personas and plan, and a startedMessage with the expected duration (usually several minutes). Results come from wait_for_results. Another batch on an existing project is run with relaunch_usability_test, which avoids a duplicate project.
list_projects
ChatGPTList the organization's projects with filtering, sorting and cursor pagination. Use to discover/disambiguate projects, resolve a project the user named, or page through a large set. Returns each project's id, title, status, type, response count and last response date, plus pageInfo { endCursor, hasNextPage, total } — pass endCursor back as cursor to fetch the next page.
org_documents
ChatGPTThe organization's own uploaded knowledge sources ("Teach AI" documents — files, links and pasted text the org added). With no query, lists those sources (name, type, status). With a query, semantically searches their CONTENT and returns a grounded answer plus supporting snippets. Use it to ground ideation or to answer questions from the org's own material.
ping
ChatGPTHealth/echo check for the TheySaid MCP connection. Returns the message you send (or "pong"). Instant — use it to confirm the server is reachable.
prepare_design_snapshot
ChatGPTCreates an upload slot for a self-contained design (HTML) so TheySaid's AI testers can run a usability test on it. Suited to designs without a durable public URL, such as a Claude Design or Artifact. Temporary preview links (for example claudeusercontent.com) expire during a test, so they are not suitable as prototypeUrl. Live apps built with v0, Lovable, Bolt or Replit are better tested through their published public URL, because a snapshot of them loses styling. The design HTML is temporarily uploaded to TheySaid. Returns { uploadId, uploadUrl, expiresInSeconds, nextStep }. Upload the HTML (text/html, CSS/JS inlined, large assets on absolute CDN URLs) with one PUT to uploadUrl, or with upload_design_snapshot when the environment blocks that host. Then: check_design_snapshot, followed by propose_usability_test(uploadId) for a new test or relaunch_usability_test(projectId, uploadId) to re-test an existing project. This tool does not start a test. Snapshots suit static designs and click-through prototypes; apps that need a live backend or login need a reachable public URL.
project_estimated_duration
ChatGPTHow long ONE project takes a respondent to complete: estimatedCompletionTimeSeconds (and the same rounded to estimatedCompletionTimeMinutes), derived from the project's question types and AI follow-up depth. A rough upper bound — branching rules may shorten the actual path. Use this when the user asks how long a survey/project takes; project_overview already includes the same figure.
project_overview
ChatGPTDeep overview of ONE project: its purpose, top themes (with sentiment, strategic recommendations and response counts), average ratings, top open-ended keywords, a few verbatim respondent quotes with share ids for citation, the project's questions (id, type, text, 1-based order, options), and estimatedCompletionTime (rough upper-bound seconds a respondent takes to finish, from question types and AI follow-up depth) — pass a question id to question_breakdown for a full per-question distribution, or to get_question for its editable detail.
project_stats
ChatGPTQuantitative metrics for ONE project: total responses, risk (negative) and growth (positive) signals, unread responses, average conversation depth, first/last response dates, and a per-channel breakdown (count + median completion). Use this for "how is project X doing" / health questions. Optional since/until ISO dates window the metrics.
propose_usability_test
ChatGPTPlans an AI usability test on a public prototype URL or an uploaded design (uploadId) without launching it: no AI testers run and nothing is spent. It creates a new draft test; re-testing an existing project is done with relaunch_usability_test instead. Plans are better with context from the user: the decision the test informs, the target users (titles, industry, company size, worldview, motivation, familiarity with the product) and the key flows. These map to learningGoals, audienceSpec and tasks. The default plan is short: about 2 tasks and 2 questions with 0-1 AI follow-ups each (a ~15-20 minute run); depth: "deep" or extra tasks/questions make it longer. Returns a proposalId, personas, tasks, questions, learning goals, an instant design scan and a reviewMessage for the user. Edits go through revise_usability_test, and the test starts only through launch_usability_test once the user approves the plan. Without a prototype source, or with a URL behind a login, the response lists the options (a public URL, or a design snapshot via prepare_design_snapshot). Fields you set (tasks, questions, audienceSpec, project, learningGoals) are applied verbatim and echoed back; omitted fields get AI defaults. Use audienceSpec (exact) or audience (free text); with personaIds, only audienceSpec.numTesters applies. voice.voiceIds takes ElevenLabs voice IDs from search_ai_tester_voices, not voice names.
question_breakdown
ChatGPTDeep results for ONE question: response count, average (rating questions), the full option distribution (option label, count, %) for multiple-choice/rating/ranking, and top phrases for open-ended. For ranking questions the distribution counts FIRST-place ranks only (each respondent contributes one), so % is the share of respondents who ranked that option top. Resolve question ids via project_overview's questions list. Optional since/until window the results.
recent_responses
ChatGPTThe most recent responses for a project (summary title, sentiment, quality, channel, share id, date). Use for 'latest feedback' or to surface citable responses.
relaunch_usability_test
ChatGPTRe-runs an existing usability-test project ("relaunch", "re-run", "run again", "same test") with a fresh batch of AI testers on the same project. Unlike propose_usability_test + launch_usability_test, it does not create a new, duplicate project. Takes projectId (the testId from an earlier launch) and an optional numTesters for this run. Without uploadId it re-runs against the design already hosted for the project; a live public URL always serves its latest version. For an updated snapshot-hosted design, pass a new uploadId (prepare_design_snapshot, then check_design_snapshot) to point the same project at the new design; the result echoes designRefreshed: true. It does not change the audience or questions; audience changes are saved first with revise_usability_test(projectId). Returns the same testId (reusedProject: true), the resolved audience and a startedMessage. Spends AI-tester runs. Results come from wait_for_results or get_test_results.
response_detail
ChatGPTFull detail of ONE response: its transcript (each message with its messageId + questionId, so you can cite a specific message), per-question answers, the themes it hit, the respondent (name/email), channel, signal, quality and date. Look it up by insightId or shareId (share ids come from citations / responses_search / recent_responses).
responses_search
ChatGPTSearch/filter individual responses across the whole org or one project. Filters: signal (positive/negative/neutral/mixed), quality (high/medium/low), channel (link/website/email/qr-code/social-media/panel), date window, free-text keyword (matches summary + respondent name/email), and a specific contactId. Returns each response's insight id, share id, summary, signal, quality, channel, date, project, and the respondent's name/email when present. Set includeConversation=true to also return each response's full conversation transcript (each turn carries its messageId + questionId, so you can cite a specific message) — use this when EXPORTING responses so the export contains the conversation. Use for 'negative responses about X', 'responses from <contact>', 'low-quality responses since <date>'.
revise_usability_test
ChatGPTEdits a usability test. Takes exactly one of proposalId (an unlaunched draft) or projectId (an already-launched test). Draft: plain-language changes and/or structured edits (add or remove tasks, questions and learning goals by index, swap personaIds, set edits.audience). Returns an updated plan under a new proposalId that replaces the previous one; nothing is launched. The number of AI testers changes only through edits.audience.numTesters. Launched test: only the audience can change (edits.audience, edits.personaIds). The change is saved to the project and echoed back as audience, and nothing runs until relaunch_usability_test. Tasks, questions and learning goals of a launched test cannot be edited; a new test is needed for that.
search_ai_tester_voices
ChatGPTSearches the curated list of AI-tester voices by name, style or accent. Returns { voices: [{ voiceId, name, gender, accent, descriptive, previewUrl }], hasMore, totalCount }. The voiceId is the value accepted by audienceSpec.voice.voiceIds in propose_usability_test and edits.audience.voice.voiceIds in revise_usability_test; voice names are not valid ids. page is 1-based; pageSize defaults to 30 (max 100).
similar_projects
ChatGPTFind projects related by SEMANTIC similarity, using each project's objective (learning-goal) embeddings — falling back to its theme embeddings when objectives are not embedded yet. Pass a seedProjectId to rank the projects most similar to it (with 0–1 scores). Pass null for seedProjectId to cluster ALL of the org's projects into thematic groups. Each project notes its basis ("objectives" or "themes"). Use this for "which projects are like X?" or "group my projects"; use cross_project_comparison instead to compare a known set side-by-side by theme.
stats_trend
ChatGPTResponse volume over time for a project, bucketed by day or week and split by signal (positive/negative/neutral). Use for trend / volume-over-time questions, then feed the points into generate_chart (line) for a visual.
theme_detail
ChatGPTDrill into ONE theme of a project: its summary, sentiment, strategic recommendations, how many responses mention it, and verbatim respondent quotes with share ids to cite. Resolve theme ids via project_overview.
upload_design_snapshot
ChatGPTFallback upload for a design snapshot, for when the PUT to the uploadUrl from prepare_design_snapshot is blocked by the environment's network (for example "Host not in allowlist"). Takes the uploadId and the complete, self-contained design HTML, and hosts it under that uploadId. The design HTML is temporarily uploaded to TheySaid. Re-using an uploadId replaces the stored design. Returns { uploadId, previewUrl, nextStep }; the next step is check_design_snapshot. This tool does not start a test.
wait_for_results
ChatGPTWaits server-side, for a bounded window, for a usability test to finish, and sends progress notifications meanwhile, so no polling is needed. Returns message (a user-facing summary), phase and, once finished, results (the same data as get_test_results, including resultsMarkdown). If the window elapses first, stillRunning is true; long runs are normal, and calling wait_for_results again continues waiting. For a "partial" or "failed" run, the partial findings and instant design scan are included. Optional maxWaitSeconds shortens a single wait.
channel_breakdown
Claudecheck_design_snapshot
Claudecompany_data
Claudecontact_detail
Claudecontact_list
Claudecross_project_comparison
Claudeflag_result
Claudegenerate_chart
Claudeget_question
Claudeget_test_results
Claudeget_test_status
Claudeinsight_video_metadata
Claudelaunch_usability_test
Claudelist_projects
Claudeorg_documents
Claudeping
Claudeprepare_design_snapshot
Claudeproject_estimated_duration
Claudeproject_overview
Claudeproject_stats
Claudepropose_usability_test
Claudequestion_breakdown
Clauderecent_responses
Clauderelaunch_usability_test
Clauderesponse_detail
Clauderesponses_search
Clauderevise_usability_test
Claudesearch_ai_tester_voices
Claudesimilar_projects
Claudestats_trend
Claudetheme_detail
Claudeupload_design_snapshot
Claudewait_for_results
Claude