Reos MCP Server for Claude and ChatGPT
By Reos
Overview
Reos is a user-research platform. Teams bring interview recordings, transcripts, documents and survey data into Reos, and Reos turns them into evidence-backed observations, insights and recommended solutions. This connector brings that workspace into Claude. With Reos connected, Claude can: - Answer questions about your research with the exact observations, insights and documents behind each claim, each linked back to Reos. - Search across projects, or read everything one participant said. - Read an interview transcript with timecodes, look at a still frame from a recording, and read the comments and markers your team left on it. - See where each project stands, from raw data to observations, insights and solutions. - Add raw data by pasting text or importing a public URL, start an analysis run and read its results when it finishes. - Create a project or save a write-up as a Reos document, and edit or remove observations and insights. How access works: you sign in to Reos and choose one organisation to connect. Claude acts as you, with the permissions your role has there, and only reaches projects you can already open. Viewers get reading tools only. Every tool that changes your workspace asks before it runs. You can revoke access at any time under AI assistants in your Reos profile menu. Reos is for researchers, designers and product teams who already keep their research in Reos and want to query and act on it without leaving Claude.
How to connect
- Claude: Open the Reos listing in the Claude connectors directory and choose Connect. Claude listing Reos exposes 25 tools here.
- ChatGPT: Open the Reos listing in the ChatGPT app directory and choose Connect. ChatGPT listing Reos exposes 25 tools here.
Tools
add_url_resource
ChatGPTtype unset unless the user said what the file is; it then defaults to 'document', which is right for PDFs, papers, articles and web pages. Use add_text_resource for pasted text. Private/internal URLs and Reos's own pages are rejected.create_document
ChatGPTtitle and content (full markdown body that you draft yourself based on the conversation, search results, and selected entities). The Document is persisted in the project's Documents list as type='custom' and the chat reply links to it. For templated kinds of document (summary, evaluation, critique, tips, analysis), use start_flow_run with outputType: "documents" and a documentTypeId instead — that runs the configured flow over project inputs and is the right path whenever a Document Type already covers what the user wants.create_project
ChatGPTdelete_entities
ChatGPTconfirmedUnselected: true to record that confirmation.edit_entity
ChatGPTcontent (and optionally title for insights and solutions) to replace the text; (2) evidence, for insights only: linkSupportingObservationIds or linkContradictingObservationIds to attach observations as supporting or contradicting evidence, unlinkObservationIds to detach them; a link list also changes the stance of an observation that is already linked; (3) tags: addTagIds and removeTagIds on any entity type, with tag ids from list_tags. Edits one item per call.fetch
ChatGPTid a search result carried - the complete observation, insight, solution, document, uploaded resource, or persona, not the snippet. Ids look like insight:<uuid>; pass one back exactly as search returned it.get_flow_run_results
ChatGPTget_flow_run_status
ChatGPTflowRunId, batchRunId or batchJobId. Poll until status is 'completed' (or 'failed' / 'cancelled'), waiting a few seconds between polls, before reading results. Returns the status, and what went wrong if it failed.get_frame
ChatGPTget_transcript segment timecodes to grab the exact moment you want to reference. Pass resourceId and timestampMs (milliseconds from the start). Set describe: true to also get a one-line AI description of what's on screen. Frames are cached, so repeat calls for the same moment are cheap.get_linked_observations
ChatGPTsupporting or contradicting). Use when the user asks what evidence backs an insight, what counters it, which observations support or contradict it, or where a finding came from. Prefer this over search_data when the user is asking about an insight that's already in the selection — the linked evidence is authoritative. Also use it before changing an evidence link's classification, so you know which observations are currently supporting vs contradicting.get_observations
ChatGPTsearch_data (semantic, at most 20 results) and list_entities (short previews), it returns full text and pages through the whole set with a cursor. Call it without a cursor first, then again with the returned cursor until it reports the end. On very large projects, read one participant at a time with participantId. For a count use list_entities; for one participant's coverage get_participant_observations is lighter.get_participant_observations
ChatGPTparticipantId or participantName: the exact number of that participant's observations, a sample of them, and which persona the participant's record is linked to, if any. Without either: every participant on the project's raw data with their exact observation count, the total number of participants, and any interviews with no participant linked. Use it for questions such as 'what do we know about this person?', 'how many observations do we have for them?' or 'how many participants do we have?'. The counts are exact whatever the sample size; search_data is semantic and capped at 20 results, so it cannot answer these.get_project_status
ChatGPTstart_flow_run, and to check that a stage has finished. researchQuestionCandidates counts questions inferred from the project's data that the user has not confirmed; they are not part of the project's research questions.get_transcript
ChatGPTstatus. Pass resourceId (get it from list_entities with entity_type 'resource', or from search results). Use the per-segment timecodes to decide where to pull a still with get_frame. status is one of: 'ready' (segments returned), 'processing' (still transcribing - try again shortly), 'no_transcript' (none exists), 'unavailable' (temporarily unreadable).list_comments
ChatGPTmarkers flags (tag, isGreatQuote, isKeyMoment); these flags are how Reos records reactions to a moment. Pass resourceId.list_documents
ChatGPTtitleContains for a case-insensitive match on the title, which finds a named document more reliably than the semantic search_data. Use it before create_document when the user names a document, to avoid creating a duplicate. Returns non-deleted documents only.list_entities
ChatGPTsearch_data is semantic, returns at most 20 results and is not a count. Results are paginated: while truncated is true, call again with the returned nextCursor and the same arguments. personas and super_insights are organisation-wide and need no project. For documents use list_documents; for an overview of every stage use get_project_status; to read every observation in full use get_observations. If a nameContains lookup finds nothing in the given project, it also reports matches in your other accessible projects.list_flows
ChatGPToutputType filter: 'observations', 'insights', 'solutions', 'documents', 'super_insights' or 'questions'. Use it before start_flow_run when you need a specific flow or want to know what the organisation has set up.list_organizations
ChatGPTorgId argument on any other tool to work in that org), slug, name, and your role there. The org flagged isCurrent: true is the one used when you omit orgId.list_projects
ChatGPTtruncated is true, more exist than shown — call again with the returned nextCursor until every project is listed. Useful when the user asks 'what projects do I have', to pick one to work on, or to establish the full project set before cross-project work.list_tags
ChatGPTedit_entity with addTagIds/removeTagIds. Pass entityType (observation/insight/solution) to narrow to tags that are valid for that entity type — otherwise returns the full catalogue.search
ChatGPTquery; the search is semantic, not keyword-based. Read a result in full by passing its id to fetch.search_data
ChatGPTstart_flow_run
ChatGPTadd_text_resource
Claudeadd_url_resource
Claudecreate_document
Claudecreate_project
Claudedelete_entities
Claudeedit_entity
Claudefetch
Claudeget_flow_run_results
Claudeget_flow_run_status
Claudeget_frame
Claudeget_linked_observations
Claudeget_observations
Claudeget_participant_observations
Claudeget_project_status
Claudeget_transcript
Claudelist_comments
Claudelist_documents
Claudelist_entities
Claudelist_flows
Claudelist_organizations
Claudelist_projects
Claudelist_tags
Claudesearch
Claudesearch_data
Claudestart_flow_run
ClaudeLinks
Similar Apps
Listed on MCP App Store
Add this badge to your site or README. It links to this page.
HTML
<a href="https://mcp-app-store.keenethics-labs.com/app/reos"><img src="https://mcp-app-store.keenethics-labs.com/badges/listed-on-mcp-app-store.svg" alt="Listed on MCP App Store" width="220" height="44"></a>
Markdown
[](https://mcp-app-store.keenethics-labs.com/app/reos)






