MCP Tools

Heretic Lazy Shot exposes every core action through a built-in Model Context Protocol server. Any MCP-compatible agent can capture, OCR, pin markers and open the beautify compositor without a mouse.

Keyword hygiene. Every capture requires a short descriptive keyword like chrome-settings or login-page. Keywords collide → auto-suffix (login-page → login-page2). Always surface the returned keyword back to the user.

Capture

Tool	When to use
`capture_display`	Whole monitor
`capture_region`	Pixel rectangle on a display
`capture_window`	Specific app by title / process fuzzy match (preferred)
`capture_active_window`	Currently focused window — risky from an agent (may grab the agent's own terminal)
`capture_tracked_window`	Pick from the recently-active window stack (needs `windowTracker` flag)

Examples

// capture_window — specific app
{ "query": "Chrome", "keyword": "chrome-settings" }

// capture_region — pixel rectangle on display 0
{ "x": 100, "y": 200, "width": 800, "height": 600, "keyword": "header-crop" }

// capture_display — full monitor 0
{ "display": 0, "keyword": "full-desktop" }

Discovery & listing

Tool	Purpose
`list_displays`	Connected monitors + geometry
`list_tracked_windows`	Recently active windows — call before `capture_tracked_window`
`list_screenshots`	Paginate gallery; filter by status / type / sort
`search_screenshots`	Query filename, keyword, metadata — with date range and type
`get_screenshot`	Full metadata for one shot (by id or keyword)
`get_app_status`	App version, MCP port, feature flags
`list_ocr_languages`	Available / installed OCR language packs
`list_marker_presets` / `search_marker_presets` / `get_marker_preset`	Saved marker overlays

// search_screenshots
{ "query": "dashboard", "date_from": "2026-04-01", "type": "window" }

Capture type values: region | window | display | beautify | markers.

Actions on an existing screenshot

Tool	Purpose
`assign_keyword`	Rename / tag — call right after capture if you want a stable name
`edit_screenshot`	Open in the Beautify compositor. `edit_existing: true` overwrites; default creates a copy
`delete_screenshot`	Soft delete
`open_screenshot`	Open in the OS viewer (Preview on macOS)
`show_window`	Show or hide the lazy-shot app window itself

// rename after capture
{ "id": "chrome-settings", "keyword": "chrome-settings-dark" }

// open in compositor (creates a copy by default)
{ "id": "chrome-settings" }

Markers — `add_markers`

Overlay numbered or labelled pins. Provide explicit markers or reference a saved preset.

{
  "screenshot_id": "chrome-settings",
  "keyword": "chrome-settings-annotated",
  "markers": [
    { "x": 120, "y": 80,  "label": "1", "color": "#FF0000", "size": 32 },
    { "x": 340, "y": 200, "label": "2", "color": "#00AA00" }
  ]
}

For reusable overlays, save a preset in the app and call by name via get_marker_preset.

OCR

Two entry points:

ocr_screenshot — run OCR on a gallery item (by id or keyword)
ocr_image_path — run OCR on any file inside the screenshots / app-data directory

Parameters that matter:

Param	Values	Notes
`format`	`text` · `metadata` · `text_and_metadata`	`metadata` returns per-word boxes + confidence
`variant`	`fast` · `best`	Use `best` only when `fast` garbles the output
`lang`	e.g. `eng`, `rus`, `eng+rus`	Non-installed packs auto-download (~4 MB each) on first call

{ "id": "login-page", "lang": "eng+rus", "format": "text_and_metadata" }

Response shape

Capture tools return a normalised object so agents can chain them:

{
  "id": "chrome-active-view-222",
  "keyword": "chrome-active-view",
  "path": "/Users/you/Pictures/lazy-shot/chrome-active-view-222.png",
  "width": 2880,
  "height": 1800,
  "type": "window"
}

Chain capture_window → add_markers → edit_screenshot → ocr_screenshot using the id or keyword from each step.

Prompt recipes

"Beautify the active Chrome tab"

Capture the Chrome window with keyword "chrome-active-view",
then open it in the beautify editor.

Under the hood: capture_window({ query: "Chrome", keyword: "chrome-active-view" }) → edit_screenshot({ id: "chrome-active-view" }).

"Capture a bug and pin three markers"

Capture the region x=0 y=0 w=1400 h=900 on display 0 with keyword "bug-42".
Pin three numbered markers at (120,80), (340,200) and (540,420).
Save the result as "bug-42-annotated" and open it.

"OCR the login page in English + Russian"

Capture the current Chrome window as "login-page", then OCR it with
lang=eng+rus and format=text_and_metadata. Return the recognised text.

"Find every dashboard shot from April"

Search screenshots: query="dashboard", type="window",
date_from="2026-04-01". Return ids and paths.

"Re-shoot using the tracked window stack"

List tracked windows. If Slack is in the list, run capture_tracked_window
on it with keyword "slack-thread". Otherwise capture Slack via capture_window.

Safety

All tools run locally — no cloud calls, no network egress
capture_active_window is risky from an agent (it may capture the agent's own terminal) — prefer capture_window with an explicit query
Keyword collisions auto-suffix (rabbit → rabbit2) — use the returned keyword in follow-up calls
Non-installed OCR language packs auto-download (~4 MB each) on first call

See integrations for how to wire Claude Desktop, Claude Code, Cursor and Copilot to this server.