accept_edit
ChatGPTAccepts a pending edit and applies it to the study, once the user has agreed to it. Applies plan, concept, market, audience, category, brand, scheduling, framework, market reach, price sensitivity, question, split and context block edits.
add_narrative_section
ChatGPTAdds a section to a narrative, after the given section or first. Changes the narrative only, not an existing deck.
add_study_note
ChatGPTRecords a note on the study's builder conversation, where the user and the study builder in the Ideally webapp read it: a decision, the reason behind a batch of proposals, or an open question.
attach_finding
ChatGPTAttaches a finding to a section as its evidence. The finding must belong to the narrative's study; find one with search_findings, or create_finding when none covers the point.
create_canvas_filter_set
ChatGPTSaves a named filter set on a Canvas: the respondents who gave the chosen answers, in the chosen segments. It shows in the Canvas's filter list for every market. Answers to one question widen it (Yes or Maybe); answers to different questions narrow it (Yes and aged 25-34). Respondents in any of the segments are kept. For age, gender and the like, use the answers to its audience's screeners. Tell the user its name and what it keeps, and call it only once they agree.
create_finding
ChatGPTCalculates a new finding from one study's or one Canvas's raw data: one question's answer shares for one audience, saved as facts the finding cites. Call it only when search_findings found nothing that covers the question, and pass studyId or canvasId: it never reads the whole account. A question or audience it cannot calculate fails with the ids it accepts. It can take a minute: it returns the finding, or a handle to pass to get_finding_status while it is still running.
create_narrative
ChatGPTStarts a new, empty narrative on a study, for a new presentation. Only when the study has none to use or the user asks for a new one; check list_narratives first.
create_study
ChatGPTCreates a draft study from the gathered brief, on the framework the user agreed to, and records the brief on its builder conversation so the study builder in the Ideally webapp sees it. It starts empty, apart from the audience of the team's only Canvas when it has exactly one: build it up afterwards with proposals the user reviews. conversation is recorded, started when no brief was given, or a warning that the brief still needs add_study_note.
create_study_filter_set
ChatGPTSaves a named filter set on a study: the respondents who gave the chosen answers. It shows in the study's filter list and in get_study_filters as a segment the results tools take. Answers to one question widen it (Yes or Maybe); answers to different questions narrow it (Yes and aged 25-34). It is saved in the market given and copied to the study's other markets, unless a demographic answer belongs to that market alone. Tell the user its name and the answers it keeps, and call it only once they agree.
delete_canvas_filter_set
ChatGPTDeletes a filter set from a Canvas, in every market. Tell the user its name, and call it only once they agree.
delete_study_filter_set
ChatGPTDeletes a filter set from a study, in one market: a copy in another market has its own id in get_study_filters for that market. The facts, findings and summaries built on it are deleted too. Tell the user its name and market and what goes with it, and call it only once they agree.
detach_finding
ChatGPTDetaches a finding from a section. The finding itself is kept.
discard_pending_edits
ChatGPTTakes back pending edits before anyone reviews them: one by its edit id, or all of them on a section, study question, context block, concept or split.
download_deck
ChatGPTBuilds the deck as a PowerPoint file and returns a download link. It builds a fresh file on every call, which usually takes a few seconds and can take up to about 20. The link expires after an hour: call again for a new one rather than storing it.
edit_narrative_section
ChatGPTChanges a section's title, body or both. Changes the narrative only, not an existing deck.
edit_slide_value
ChatGPTReplaces one text field on a slide, by the name get_deck lists under text. For anything else (charts, layout, images, a rewrite) use revise_deck.
edit_speaker_notes
ChatGPTReplaces a slide's speaker notes with plain text, or clears them with an empty string. Nothing on the slide changes.
generate_deck
ChatGPTGenerates a presentation deck from a narrative, once per narrative. Shape the sections and attach findings first. It takes 5 to 10 minutes: it returns the status and a handle for get_deck_status.
get_canvas
ChatGPTGets one Canvas: its name, category, team, the markets it covers and its runs by date. Use the markets and run dates it lists for the other Canvas tools.
get_canvas_datasets
ChatGPTGets what a Canvas owns as Datasets: canvas_studies (its U&A studies, which are its waves, with their fieldwork dates; any other study in the Canvas is read only through its own share), canvas_runs (each of those studies' runs by market, and whether each completed), segments (its published segmentation, the one get_segments describes), segment_membership (which respondent sits in which segment), its questions (canvas_questions, canvas_question_options, canvas_question_routings) and its demand zones (demand_zones, demand_zone_cells). Respondent rows stay in each study's own Datasets: get them with get_study_datasets and join on study_id and respondent_id in run_dataset_sql, for example to compare waves by segment. segments and segment_membership are empty until the Canvas has a published segmentation. The first call for a Canvas builds its data: it returns the Datasets, or a running status while the build continues, and calling it again picks the same build up.
get_canvas_questions
ChatGPTLists a Canvas's questions in one market, a page at a time: each question's id, title, text, type and answer options. Use the ids for get_crosstab.
get_chart_evidence
ChatGPTGets the figures behind a finding search_findings returned, ready to chart: each question's series by audience, with bases, significance, the total and benchmarks. Draw the chart from these numbers as given.
get_crosstab
ChatGPTCrosstabs a Canvas's questions in one market by its segments, or by the answers to another question, a page of questions at a time: pass the same questionIds, by, byQuestionId, market and runDate with the NextCursor for the rest. Each question has its base (how many answered it), and each answer its share among all respondents (total) and, per column, the count, the column's base, the percentage, an index against the total (100 is the same as the total) and whether the difference is significantly higher or lower.
get_dataset_sql_status
ChatGPTFollows a run_dataset_sql that was still running: returns its rows once it finishes, or the same handle while it is still running.
get_deck
ChatGPTGets a narrative's deck as an outline: each slide's slotId, title, editable text fields, other content, speaker notes and cited findings, plus the checkpoints undo and restore_checkpoint move between. Address slides by slotId, never by number.
get_deck_status
ChatGPTSays whether a narrative's deck is none, generating, ready or failed, and how long it has been generating. Follow up a generate_deck that was still running.
get_demand_landscape
ChatGPTGets a Canvas's demand landscape in one market: its demand zones, largest first, each with its share of the market's occasions and the occasion-by-segment cells it is made of. Cells outside every zone are grouped last. It counts the Canvas's U&A studies and the studies shared with the Canvas's owner team, the same for every caller.
get_drivers
ChatGPTGets the data behind a study's driver chart for one open-ended question, one row per theme for all respondents or per concept. Each row gives the theme's impact on each metric with its p-value and whether it is significant (p below 0.05), how many responses mention the theme (as the dashboard counts them) and its sentiment. Impact is a regression coefficient in the metric's answer-scale points: how far a respondent's score moves when they mention the theme, other themes held constant; negative drags the metric down. Drivers are study-wide, never per segment. Themes whose absolute impact on every metric is below 0.05 are left out, strongest first. available is false when the driver model has not scored the question. Use it to chart, rank or compare drivers; use query_study_facts with driver_result to read them as written sentences.
get_finding_status
ChatGPTFollows a create_finding that was still running: returns the finding once it is written, or the same handle while it is still running.
get_finding_verbatims
ChatGPTGets respondents' own words behind a finding search_findings returned, exactly as stored, with response ids. Never invent or paraphrase a quote. If the finding does not cover the audience, concept or theme asked about, search for one that does first.
get_narrative
ChatGPTGets a narrative: its brief, and its sections in order with the findings each one cites. Address sections by id, never by position.
get_previous_studies_summary
ChatGPTSummarises the ten most recent completed studies: framework, category, brand, markets, concepts and the questions the team wrote. For what one cost, use get_study_cost.
get_price_sensitivity
ChatGPTGets a study's price sensitivity (Van Westendorp) results: the optimal price, the indifference price, the acceptable range, and the curve of who finds each price too cheap, a bargain, getting expensive or too expensive.
get_reference_sql
ChatGPTGets Ideally's reference SQL over raw data, which Ideally's tests prove reproduces the dashboard's numbers (raw-data) and the stored Facts of a study or Canvas (facts-sql). Without arguments it lists the guides and their queries. With guide it returns that guide: the tables, the rules for weights, bases and rounding, and what each query takes. With query it returns that query's SQL, naming its tables as {{table}} and its inputs as {{parameter}}: replace each {{table}} with ds_<datasetId> from get_study_datasets or get_canvas_datasets and each {{parameter}} with its value, change only what the question needs, and run it with run_dataset_sql. Start from the closest query rather than writing SQL from scratch.
get_research_plans
ChatGPTGets the research plan of each study: the background, what it set out to learn, its type, framework and audience. Read it once the studies are chosen, to frame the answer. A plan is intent, never evidence.
get_revision_status
ChatGPTFollows a revise_deck that was still working: returns the deck agent's reply once the revision is done.
get_segments
ChatGPTGets a Canvas's published Segmentation: each segment's name, the rules that put a respondent in it as readable text, and how many respondents it holds. A Canvas with no published Segmentation has no segments.
get_study
ChatGPTGets one study's plan: status, Canvas, framework, category, brand, background, objectives, dates, its schedule (start date, waves and weeks between them), marketReach (the market reach add-on with its maxOptionsSelect, left out when the study has none), and studyUrl, its link in the Ideally webapp.
get_study_conversation
ChatGPTReads the study's builder conversation, oldest first: what the user, the study builder in the Ideally webapp and MCP clients said about the study. Each message says who it is from: user, webapp_builder or mcp. Read it before working on a study that already exists.
get_study_cost
ChatGPTGets what a study costs in credits: credits already charged and still to charge, per market, and the team's and organisation's credits left after those charges.
get_study_datasets
ChatGPTGets a study's raw, respondent-level data as Datasets: one table each, with its columns and row count. They cover every region and run, so they grow large; query them with run_dataset_sql and aggregate in SQL. Use them for a cut or calculation no results tool answers in one call: combined conditions, quotes and themes from one sub-group, crosstabs between questions. The first call for a study builds its data: it returns the Datasets, or a running status while the build continues, and calling it again picks the same build up.
get_study_design
ChatGPTGets one study's survey design a page of questions at a time, in the order respondents see them, section by section: each question's type, answer options (an option with excludedFromBase true, such as a standard metric's "Not sure", is shown to respondents but its answers count in no base), the concept it belongs to, the block it sits in and its routing as readable text. Each page also carries the context blocks shown among its questions, and the first page lists the study's concepts and its splits with their blocks. Each question has two ids: studyQuestionId is its place in this study, which the results, driver and edit tools take; questionId is the question itself, shared by its copies when it is asked once per concept, and is what templates, Facts' keys.questionId and the raw data's question_id use.
get_study_filters
ChatGPTLists what a study's results can be cut by: the demographics (e.g. Gender, Generation), with each answer's id and how many respondents gave it, and the study's saved segments. Pass the answer ids as demographics, or a segment id as segment, to get_study_results, get_verbatims and the other results tools.
get_study_results
ChatGPTGets a study's results per concept, the way the dashboard shows them. A metric gives its top-two-box score and base (n) against the Ideally average norm, with the difference in points and whether it is significant, and against all respondents when a segment is chosen. Any other question gives each answer's share, count and base. Without studyQuestionId it covers the study's metrics; pass studyQuestionId for any other question.
get_turf
ChatGPTGets a study's TURF (market reach) results: how many people each option reaches on its own, and the best combinations of combinationSize options ranked by reach, with frequency and people reached. Exclusive options such as "None of the above" are left out, as the dashboard does.
get_verbatims
ChatGPTGets what respondents wrote to one open-ended question, a page at a time, with the concept each answer is about and the themes it was coded to. Later pages keep the first page's limit.
list_audiences
ChatGPTLists the audiences available to this account a page at a time: each one's category, whether it screens respondents to people active in that category, and the Canvases built on it. Narrow by category or words before paging through them all.
list_benchmarks
ChatGPTLists the norms a study's metrics can be compared with in one market: Ideally, team, category and product type averages, medians and top 25%, each with its top-two-box score and roughly how many studies it rests on.
list_canvas_filter_sets
ChatGPTLists the filter sets saved on a Canvas, by name, with their ids. They show in the Canvas's filter list for every market.
list_canvases
ChatGPTLists the Canvases this account can see, newest first, a page at a time.
list_frameworks
ChatGPTLists the research frameworks a study can be built on, by study type, with what each is for, examples, and the standard questions it adds to every study on it. Recommend one from the brief before create_study. For a standard question's wording and scale, use list_standard_questions.
list_narratives
ChatGPTLists a study's narratives, newest first. A narrative is the story a presentation deck is built from. deck says whether it has a deck: none, generating or ready.
list_pending_edits
ChatGPTLists the proposed edits on a study that nobody has accepted or rejected yet, oldest first. Each says who proposed it, mcp or webapp_builder (the study builder in the Ideally webapp), and its clientApp names the app, such as Claude Code.
list_question_templates
ChatGPTFinds the saved question templates a study can reuse as they are, by section and words, a page at a time. Each says how many options it has; ask for the options, with each screener option marked as qualifying or screening out, with includeOptions.
list_regions
ChatGPTLists the markets a study can run in, with their currency and default sample size.
list_skills
ChatGPTLists the Ideally skills: guides for doing a task with these tools.
list_standard_questions
ChatGPTLists Ideally's standard questions: each one's wording, its answer scale in order, and the frameworks that add it. These are Ideally's scales, such as purchase intent, appeal and likeability: read one before writing or checking a question that should use it. positive marks the options a score counts; excludedFromScores marks a don't-know option. With a study, the wording and scales are the ones that study gets. A question some frameworks word differently is worded as the framework asked for, else the study's, else the first that adds it.
list_studies
ChatGPTLists the studies this account can see, most recently active first, a page at a time.
list_teams
ChatGPTLists the teams under this account that the user can act in, with their role on each and whether they can create studies there. A study belongs to one team, so when there is more than one, ask the user which.
load_skill
ChatGPTLoads one Ideally skill's full guide. Read it before starting that task. With a query, returns the SQL of one of the reference queries the skill names.
move_narrative_section
ChatGPTMoves a section one place up or down the running order. Call it again to move further.
org_switcher
ChatGPTSwitches this connection into another organisation. The user is asked to sign in and approve the switch in their browser, so tell them first. Most apps show this as a Forbidden, insufficient scope or authentication error with a Connect button: that is the approval step, not a failure, so tell the user to click Connect and approve, and never report it as a bug. If no Connect button appears, tell them to disconnect and reconnect the Ideally connector and choose the organisation. Afterwards every tool works in that organisation. Get the clientId from search_orgs.
propose_audience
ChatGPTProposes the study's audience, from list_audiences, or a Canvas whose audience it takes.
propose_brand
ChatGPTProposes whether a study is branded, and the brand.
propose_category
ChatGPTProposes the study's category and product type.
propose_concept_edit
ChatGPTProposes adding, editing or removing one of a study's concepts.
propose_context_block_edit
ChatGPTProposes adding, editing or removing a context block shown to respondents.
propose_framework_change
ChatGPTProposes moving a study to another framework from list_frameworks.
propose_grid_edit
ChatGPTProposes adding, editing or removing a grid: statements shown as the rows of one table, sharing one scale. Add a grid in one step with statements, header, questionType, options and position, or from saved questions with studyQuestionIds. Removing a grid keeps its rows as separate questions. Screeners from the study's audience can't be rows, and a grid of them can't be changed or removed here. A Canvas's seed study takes grids in its OutsideLoop (U&A) questions only. List the options best first, such as Strongly agree to Strongly disagree.
propose_market_reach
ChatGPTProposes adding, changing or removing the market reach (TURF) add-on.
propose_markets
ChatGPTProposes the markets a study runs in. Each panel you list markets for (the Ideally panel, or the team's own panel) is replaced by those markets; a panel you list nothing for keeps its current markets. Region ids come from list_regions.
propose_plan_edit
ChatGPTProposes a new background or research objectives for a study.
propose_price_sensitivity
ChatGPTProposes adding, editing or removing price sensitivity in one market. Price sensitivity runs in one market only: a proposal for another market is refused while one market has price sensitivity saved or pending, and the refusal names that market. A second proposal for the same market updates its pending edit.
propose_question_edit
ChatGPTProposes adding, editing or removing a survey question: its text, type, options, routing, the concept it is asked for, and the block it sits in. A saved template from list_question_templates can be added as it is.
propose_scheduling
ChatGPTProposes when and how often a study runs. Only Ask Anything and In-market test studies have scheduling. On any other framework, such as Concept Test, the study runs as soon as it's launched in the webapp, so do not propose a schedule.
propose_split_edit
ChatGPTProposes adding, editing or removing a split: a set of blocks of questions, of which each respondent sees some or all.
query_canvas_facts
ChatGPTReads a Canvas's facts: data points Ideally computed across its studies, each a sentence with its base (n), whether the base is low (under 30), its index against the category total, and the keys that match it to a row of its ideally/facts-sql query. Pass the factIds get_chart_evidence returns to read the facts a Canvas finding rests on; otherwise it lists one market's facts, newest run first. Use search_findings to search them by topic.
query_study_facts
ChatGPTSearches a study's facts: computed data points from its results, each a sentence with its numbers, base (n) and comparisons, best matches first. Use it to find what the data says about a topic before reading full results.
redo
ChatGPTSteps the deck forward again after an undo.
reject_edit
ChatGPTRejects a pending edit, leaving the study as it is.
remove_narrative_section
ChatGPTDeletes a section and its finding links. Only when the user asks for it to go. The findings themselves are kept.
remove_slide
ChatGPTDeletes a slide from the deck. Only when the user asks for it. undo brings it back.
reorder_slides
ChatGPTPuts the slides in a new order. Pass every slotId exactly once.
restore_checkpoint
ChatGPTPuts every slide back as it was at a checkpoint from get_deck. The restore is itself a checkpoint, so undo reverses it. Name the checkpoint to the user and ask before a jump of more than one step.
revise_deck
ChatGPTAsks the deck agent to re-author the deck: rewrite or restyle slides, change a chart, insert a slide, apply narrative changes. Use the direct edit tools for notes, text fields, order and removals. The instruction and reply appear in the webapp's deck chat. It can take a few minutes: it returns the reply, or a handle for get_revision_status.
run_dataset_sql
ChatGPTRuns one DuckDB SELECT you write over Datasets from get_study_datasets or get_canvas_datasets, to compute an answer from a study's raw data. A study's data covers every region and every run, so respondents add up to many thousands and answers to far more, but at most 500 rows and 50 columns come back (fewer when the result is wide, and about 2 MB of text at most, with any text value cut at 4 KB), so do the work in SQL: GROUP BY, count(), sum(weight), averages and percentages, and return the summary, not respondent rows. Never fetch raw rows to count or average them yourself; a truncated result is a sample of the first rows, not the data. Read each Dataset as ds_<datasetId>, for example ds_7. Only those tables, your own CTEs and the unnest, range and generate_series functions can be read, and PIVOT is not supported: pivot with count() FILTER (WHERE ...) or CASE. It returns the rows, or a handle to pass to get_dataset_sql_status while it is still running. A refused or failing query returns the reason, with DuckDB's own message, so fix the SQL and run it again. Report a figure computed here as computed from the raw data, never as a finding, and give it no finding citation.
search_canvases
ChatGPTFinds the Canvases this account can see by words in their name or category, newest first, with their run dates and whether they have findings to search. Use it to choose which Canvases a question is about, and ask the user when the fit is not clear.
search_findings
ChatGPTSearches the account's research findings for ones that answer a question. Call it first for any question about the account's data. Each finding has a ref to cite it by; a finding listed under otherBenchmarks is the same result against another benchmark and needs its own citation. hasQuotes says whether get_finding_verbatims has respondent quotes for it, so for quotes pick a finding that has them. Unless the query names an audience, whole-sample findings rank above audience cuts. Without studyIds or canvasIds it searches every study and Canvas with findings.
search_orgs
ChatGPTFinds the organisations the user could switch this connection into, by words in their name and despite typos, best match first, and says which one the connection is in now. Use it to get the clientId for org_switcher.
search_studies
ChatGPTFinds the studies this account can see by words in their name, category or research objectives, newest first, with their objectives, run dates and whether they have findings to search. Use it to choose which studies a question is about, and ask the user when the fit is not clear.
undo
ChatGPTSteps the deck back one checkpoint, whoever made the change: this client or someone in the webapp. Slides only; narrative edits stay.
whoami
ChatGPTReturns the identity this MCP session is authenticated as.
accept_edit
Claudeadd_document
Claudeadd_narrative_section
Claudeadd_study_note
Claudeask_dataset
Claudeattach_finding
Claudecreate_finding
Claudecreate_narrative
Claudecreate_study
Claudedetach_finding
Claudediscard_pending_edits
Claudedownload_deck
Claudeedit_narrative_section
Claudeedit_slide_value
Claudeedit_speaker_notes
Claudefinish_document_upload
Claudegenerate_deck
Claudeget_brand_context
Claudeget_canvas
Claudeget_canvas_datasets
Claudeget_canvas_questions
Claudeget_chart_evidence
Claudeget_crosstab
Claudeget_dataset_answer_status
Claudeget_dataset_sql_status
Claudeget_deck
Claudeget_deck_status
Claudeget_demand_landscape
Claudeget_document_search_status
Claudeget_drivers
Claudeget_finding_status
Claudeget_finding_verbatims
Claudeget_narrative
Claudeget_previous_studies_summary
Claudeget_price_sensitivity
Claudeget_research_plans
Claudeget_revision_status
Claudeget_segments
Claudeget_study
Claudeget_study_conversation
Claudeget_study_cost
Claudeget_study_datasets
Claudeget_study_design
Claudeget_study_filters
Claudeget_study_results
Claudeget_turf
Claudeget_verbatims
Claudelist_audiences
Claudelist_benchmarks
Claudelist_canvases
Claudelist_documents
Claudelist_frameworks
Claudelist_narratives
Claudelist_pending_edits
Claudelist_question_templates
Claudelist_regions
Claudelist_skills
Claudelist_standard_questions
Claudelist_studies
Claudelist_teams
Claudeload_skill
Claudemove_narrative_section
Claudeorg_switcher
Claudepropose_audience
Claudepropose_brand
Claudepropose_category
Claudepropose_concept_edit
Claudepropose_context_block_edit
Claudepropose_framework_change
Claudepropose_market_reach
Claudepropose_markets
Claudepropose_plan_edit
Claudepropose_price_sensitivity
Claudepropose_question_edit
Claudepropose_scheduling
Claudepropose_split_edit
Claudequery_canvas_facts
Claudequery_study_facts
Clauderedo
Claudereject_edit
Clauderemove_narrative_section
Clauderemove_slide
Claudereorder_slides
Claudereport_bug
Clauderestore_checkpoint
Clauderevise_deck
Clauderun_dataset_sql
Claudesearch_canvases
Claudesearch_documents
Claudesearch_findings
Claudesearch_orgs
Claudesearch_studies
Claudestart_document_upload
Claudeundo
Claudewhoami
Claude