prophecy_add_connection
ChatGPTfabricId (from prophecy_fabrics); kind preselects a connector. The widget saves it itself — no further tool call, and secrets never travel as tool arguments.By Prophecy
Your business context. Prophecy's governed data work. Answers you can check and rerun, in Claude. You know which accounts matter, what "on track" means and where the numbers break. What slows you down is the data work behind the answer: pulling, cleaning and joining messy data across systems. Prophecy DataGPT does that work end to end, from raw data to finished analysis and reports, without leaving the conversation. Ask "Which regions are behind on pipeline this quarter?" DataGPT pulls CRM exports and warehouse tables, cleans and matches records, applies your rules, and returns the answer with the visual workflow behind it. Inspect any step, open the code, or say what to change in plain language. For business teams: Go from raw data to a trusted report in plain language, using your team's definitions Turn messy, multi-tab Excel workbooks into clean tables on your data platform Check any number: every answer comes with the workflow and code behind it Rerun approved workflows on new data the same way every time, or schedule them For data teams and IT: Review and approve AI-built workflows instead of rebuilding them, backed by open code in Git Run everything on the Snowflake, Databricks or BigQuery you already own, with no separate engine Keep business-built work governed: traceable to source, separated by team, access by invite or domain Grow a knowledge graph from your catalogs that learns from every approved workflow About Prophecy: since 2017, Prophecy has built governed, code-backed data workflows for the world's largest enterprises. DataGPT brings that engine into the AI tools your teams already use. Try asking: "Reconcile this month's invoice export against payments in Snowflake and flag mismatches." "Count anything within $1 as a match, rerun, and share the unmatched list with my AP manager." Get started free: sign up for a Prophecy account when you connect. If your company manages connectors, your admin can add it for everyone.
fabricId (from prophecy_fabrics); kind preselects a connector. The widget saves it itself — no further tool call, and secrets never travel as tool arguments.pipeline / app. • QUESTIONS / CHARTS about a data file in the Databox (csv, json, parquet, txt), whether uploaded now or already there: pass the ask plus every Databox file_path it is about, both in message and in files; Totals, rankings and charts about that file come from here, computed on their fabric. ❌ WRONG: a total, count or description of that file's contents answered without calling this tool. A workbook (.xlsx/.xlsm) goes to prophecy_analyze_excel instead — Prophecy AI cannot analyze one. • Diagnostic Q&A ('why is column Y null?', 'trace lineage of Z'). Skip it for RUN-only (prophecy_run_pipeline), INSPECT-only (prophecy_pipeline_canvas, prophecy_run_table), chart audits (prophecy_audit_chart), cloud-file reads (SharePoint, OneDrive, S3, GCS, SFTP, Smartsheet, Databricks Volumes → prophecy_create_pipeline, then prophecy_analyze_excel), deletes (prophecy_delete_pipeline / prophecy_delete_analysis), and SCHEDULE intents (prophecy_pipeline_schedules / prophecy_schedule_pipeline, then prophecy_publish_project to make them run). Then: pipeline create / edit → prophecy_pipeline_canvas; if the ask includes a run → prophecy_run_pipeline INSTEAD (it draws the canvas; Prophecy AI's own run does not stream); analysis create / edit → prophecy_analysis. No project yet: create one with prophecy_create_project, then prophecy_create_pipeline in it, and tell the user which project you created. One running turn per project. Returns {conversationId, turnId, status:'running'} at once; call prophecy_ai_wait with those ids until status:done.prophecy_ai turn. Long-polls and returns status:done / status:error with the answer, or a still-running / awaiting-the-user's-answer status (call again).prophecy_ai turn that creates or edits an analysis (its own run does not stream). Pass appName + its pipeline.prophecy_databox_upload_file, then pass the returned file_path with connection: 'ProphecyFileSystem'. A cloud file: its path plus the connection name. An existing gem for this file + connection is reused.fabricId is required: take it from prophecy_fabrics (the active workspace default, unless the user named another), and never guess one — ask the user when it is unclear.chart is what the chart IS and datum the point clicked. Route chart audits here, not to prophecy_ai.prophecy_ai, app set to this name, and show it with prophecy_analysis.prophecy_ai, pipeline set to this name.processId is the gem's id or canvas label (from prophecy_analyze_excel or prophecy_pipeline_canvas); omit ports for all of them. Rows come from the gem's capped interim (default 10 MB), so a large output lands as a sample — repeat the reply's row counts and truncation flags. Not how output ports are created: prophecy_analyze_excel and the extract widget persist those. Requires confirm:true.path and renders it as a browsable widget; pass no path for the Home landing. Every call mounts another browser card, so call it ONCE per turn, for the one folder the user wants to look at: when you need folder contents to answer in text — naming files back, counting, recursing, checking something exists — use prophecy_databox_list instead, which renders nothing and takes several folders at once. A project's files may sit at ['projects', <projectId>] or, if uploaded earlier, in its attachments child — list both before concluding a file is not there.from → to, both as prophecy_databox returns an item's path. The reply is text only, naming where the copy landed; call prophecy_databox_list on the folder if you need its contents. Fails if something sits at to unless overwrite is true — ask the user before setting it. A file keeps its extension: samples.json → report.json, never → samples. A connection path cannot be written to; some stores cannot copy and say so.path (segments as prophecy_databox takes them; [] is Home). A connection path cannot be written to.path: moves it to the Databox Trash, where the user can restore it. Deleting an entry already in the Trash (a ['trash', …] path) removes it for good. Requires confirm:true (ask the user first). A connection path cannot be deleted.path (segments as prophecy_databox returns them); the token in the URL is the credential. Only when the user asks to download or keep the file: give them the link, never fetch it yourself to read, total or chart the data (questions about a file → prophecy_ai). Not for connection paths.prophecy_databox is only for when the user wants to SEE the browser, and each of its calls mounts another file-browser card. Pass every folder you need in ONE paths call rather than one call per folder, and never list a folder whose entries an earlier result already gave you.from → to, both as prophecy_databox returns an item's path. Destination folders that don't exist yet are created, so don't create them first. The reply is text only, naming where the file landed; call prophecy_databox_list on the folder if you need its contents. Fails if something sits at to unless overwrite is true — ask the user before setting it. A file keeps its extension: samples.json → report.json, never → samples, so don't offer that. A connection path cannot be written to.fabricId the prophecy_databox result carried; otherwise choose one as the fabricId description says. Bytes or a link to keep → prophecy_databox_download_file; browsing → prophecy_databox.path. If your host can hand over attachments, pass it as file and this server fetches it and returns the saved file_path; otherwise the reply returns a signed URL — POST the raw file bytes there as it shows and its 2xx response returns the file_path. ['projects', <projectId>] for a project's files; for personal ones, the user's own folder as prophecy_databox reports it for Home — the server names that folder, so a path you assembled from a prophecy_whoami userId may address one that does not exist. A connection is not an upload destination, so tell the user before calling. Several files → ONE call: hand them all over in files, or, POSTing bytes yourself, set count to the number of files and POST each to its own URL from the reply, all at once — never one call per file. Max 4 GB a file. If you hold an attachment neither as file nor as bytes, or its transfer fails, the chat dropped it; attaching it again fails the same way: tell the user the file is too large for the chat to hand over, and call prophecy_databox so they drop it into the Databox there. Then: no work named ('save this', a bare attachment) → prophecy_databox on its folder, so they see it. Any question, total, ranking, analysis or chart is work — 'add these to my files AND analyse …' is a work ask. With no project yet, create one with prophecy_create_project first and upload into ['projects', <projectId>]. A spreadsheet (.xlsx/.xlsm) → prophecy_analyze_excel with the returned file_path and connection: 'ProphecyFileSystem'. Any other format (csv, json, parquet, txt) → prophecy_ai with the ask verbatim plus the file_path, that same path also in files, then prophecy_analysis for the charts. The resulting total, aggregate or chart lands on the user's fabric, where their pipelines can reuse it.force. Show the user which pipeline and call once they agree (confirm:true).prophecy_analyze_excel already added, when the user clicks a sheet tab in the extract widget.connections: null; a per-fabric failure is an error on its row.prophecy_run_gem).prophecy_ai_wait for the answer.prophecy_audit_chart.name into prophecy_list_databases. For every fabric's connections at once use prophecy_fabrics.prophecy_run_pipeline draws this same canvas.prophecy_ai. A file the user attached, pasted or pointed at is DATA in EVERY format (csv / json / txt too) → prophecy_databox_upload_file, unless the user explicitly asked for the project source or git ('commit this', 'put it in the repo'); never infer that. Show the path and content and call once they agree (confirm:true).prophecy_ai writes a file so the user sees what landed. Binary files come back empty; user files in the Databox are read with prophecy_databox_download_file.prophecy_ai. Show the user the path and new content and call once they agree (confirm:true).confirm to get the list of schedules each fabric will add, change or remove (nothing is written); show that to the user; then call again with confirm:true and the impactKey the preview returned. The confirm call saves uncommitted changes first, and returns a new list instead of publishing if the changes moved. Simple version control projects only._meta.authorization and no content. Do NOT call from the assistant.prophecy_ai. EDIT + RUN → prophecy_ai, then this (Prophecy AI's own run does not stream). Whole pipelines only; a single gem is prophecy_run_gem.fabricId + connectionName + database + schema + table. Pass `connectionName` whenever the conversation names it (a Databox path connection/<name>/…, a clicked card), or the read uses the fabric's DEFAULT connection. Off the default connection database and schema are required, and where / orderBy / limit / columns / COUNT() do not apply. Free-form mode*: sql (SELECT-only, explicit LIMIT) for JOINs or aggregations, always on the default connection. A Source gem in a pipeline → prophecy_fast_run_gem; columns only → prophecy_describe_table.prophecy_on_the_fly_analysis) as a real pipeline + SQL App. No side effects: returns structured.nextActions[0] = { tool: "prophecy_ai", args } to fire next with args.message verbatim, then check structured.expectedCharts with prophecy_analysis. No projectId yet → prophecy_create_project first.enabled:false to pause it). File-arrival schedules are edited in the Prophecy UI. The schedule is saved to the project but does NOT run until prophecy_publish_project publishes and deploys it - say so to the user.Add this badge to your site or README. It links to this page.
HTML
<a href="https://mcp-app-store.keenethics-labs.com/app/prophecy-datagpt"><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/prophecy-datagpt)