markupR

February 15, 2026

Context-aware capture. Agent-ready handoff.

markupR v2.6.8 makes desktop and MCP capture more explicit for AI agents: each shot can carry cursor position, active app/window hints, and focused-element context when available.

The problem

Your agent still misses intent from raw screenshots.

Even with a transcript and images, agents can misread what you meant: where your pointer was, which window was active, and what element you were referencing are often missing.

That ambiguity creates extra back-and-forth and slower fixes, especially when you're reviewing dense interfaces or multi-window workflows.

The solution

Capture context, not just pixels.

v2.6.8 enriches capture events with cursor coordinates, active app/window metadata, and focused-element hints (best effort). Reports now carry the same contextual signals you used while narrating.

This metadata flows through desktop app output, CLI templates, and MCP sessions so your agent sees what changed, where attention was focused, and what to fix next.

931 tests. Same MIT license. Same pipeline, now with stronger context fidelity.

What's new in v2.6.8

Five upgrades for precise agent handoff.

Context metadata now travels with your captures across desktop, CLI, and MCP workflows.

feedback-report.md
### FB-003: Primary CTA disabled state unreadable
The button text almost disappears in dark mode.
Looks like contrast token fallback is wrong.

![Screenshot at 01:42](screenshots/fb-003.png)
Capture context: cursor 1186,642 • app "Arc" • window "Settings"
Focus hint: button[aria-label="Retest"]
Cursor-aware captures Shot markers now carry cursor coordinates so your agent can reason about precise screen location.
Active window metadata Capture context includes frontmost app/window hints for multi-window and multi-app sessions.
Focused-element hints Focused element role/text/title hints are included when available to reduce ambiguity in UI feedback.
MCP parity MCP screenshot and recording flows now persist capture context in session metadata, not just image files.
UI reliability polish Dark-mode BYOK readability and readiness-sync issues are fixed for a cleaner first-run experience.

All interfaces still use one pipeline: Capture → Transcribe → Align → Enrich → Generate. 931 tests across the suite.

Four interfaces, one pipeline

Use it however you work.

Desktop App

Menu bar → ⌘⇧F

Lives in your macOS menu bar. Hit the hotkey, narrate what you see, stop, paste. Full GUI with session history and annotation tools.

CLI

npx markupr analyze ./recording.mov

Same pipeline, no desktop app required. Watch mode monitors directories. Point it at a video or let it process recordings as they appear.

MCP Server

npx --package markupr markupr-mcp

Your AI agent triggers captures mid-conversation. No copy-paste, no context switching. The agent asks to see your screen, markupR delivers the report.

GitHub Action

eddiesanjuan/markupr-action@v1

Add visual feedback to every PR in CI/CD. The action processes recordings and posts structured reports as PR comments automatically.

Under the hood

The pipeline.

Every interface -- desktop, CLI, MCP, GitHub Action -- runs the same core pipeline. Capture → Transcribe → Align → Enrich → Generate.

  • Context-aware capture snapshots Shot events now include cursor coordinates, active app/window hints, and focused-element context (best effort) for clearer agent interpretation.
  • Desktop + MCP parity Context metadata now flows through desktop output, MCP session metadata, and generated templates so every interface tells the same story.
  • Template-aware report output Structured Markdown and JSON outputs now carry capture context near relevant frames to reduce follow-up clarification.
  • BYOK reliability polish Dark mode readability and key-readiness sync were tightened so onboarding and advanced settings stay clear and accurate.
  • 931 tests, MIT licensed No telemetry. No tracking. No analytics. Open source with comprehensive test coverage across the full pipeline.

Get started

Pick your interface.

Desktop App

Download the macOS or Windows app. Lives in your menu bar. BYOK keys are optional depending on your transcription/analysis path.

macOS install note: if first launch shows an unidentified-developer warning, use right-click → Open once.

Download

CLI + Watch Mode

Process recordings from your terminal, or watch a directory for new files.

$ npx markupr watch ./recordings

MCP Server

Add to your AI agent's config. Copy the JSON below.

$ npx --package markupr markupr-mcp

GitHub Action

Add to your CI/CD workflow. Visual feedback on every PR.

uses: eddiesanjuan/markupr-action@v1