Skip to content

Copilot + MCP dashboard and query workflow in extension

Problem

Analysts can invoke Dataface MCP tools from Copilot/Cursor chat in raw form, but the experience is fragile and disconnected: generated YAML doesn't automatically land in the right file, preview doesn't refresh after MCP output, and there are no guardrails preventing Copilot from producing invalid dashboard definitions. The round-trip from "ask Copilot to build a chart" to "see it rendered in preview" requires multiple manual copy-paste and save steps. Until the extension wires MCP tool outputs into the file system and preview lifecycle as first-class commands, the AI-assisted authoring workflow remains a demo, not a usable pilot feature.

Context

  • Analysts can invoke MCP-backed actions from Copilot context and receive usable outputs.
  • Dashboard/query outputs round-trip into files and preview without manual glue steps.
  • Guardrails and prompts prevent common failure modes during pilot sessions.

Possible Solutions

Plan

  • Define extension command surface and prompts for Copilot+MCP authoring actions.
  • Implement command wiring and state sync between Copilot output, files, and preview.
  • Validate with pilot scenarios (new dashboard, query correction, chart edit).
  • Document supported/unsupported Copilot flows for pilot participants.

Implementation Progress

Deferred from M1 → M2

Reason: This requires wiring MCP tools into Copilot authoring loops with round-trip file sync, preview integration, and a designed command surface — a large unsolved integration problem. The MCP tool contract isn't finalized, and the underlying capability already works in raw form: analysts can use the MCP server directly from Copilot/Cursor chat without custom extension commands. Polishing this into a seamless extension experience is M2 work once the MCP contract stabilizes and pilot feedback identifies the highest-value Copilot workflows.

  • mcp-analyst-agent M1 tooling contract task

Review Feedback

  • Review cleared