menghuaban520.github.io

read_capture_text

Read text from one saved capture using the explicitly local macOS OCR helper. Does not read a live tab or click. Images stay local; recognized text is exposed to the agent host. Read once, then reuse exact text_id and image_hash with focus_capture_text. Reliable spans also expose a content-versioned ui_target test ID for the visible image-text button, so a capable host can avoid repeated tool round trips. Missing or stale UI targets require reopening/rereading, never coordinate guessing. Untrusted OCR may be wrong; scores are not calibrated probabilities. Requires the local dev helper; no remote fallback.

Page: /trace-webmcp/

Input schema

{
  "$schema": "https://json-schema.org/draft/2020-12/schema",
  "type": "object",
  "properties": {
    "evidence_id": {
      "type": "string",
      "minLength": 1,
      "maxLength": 100
    },
    "offset": {
      "default": 0,
      "type": "integer",
      "minimum": 0,
      "maximum": 5000
    },
    "limit": {
      "default": 80,
      "type": "integer",
      "minimum": 1,
      "maximum": 200
    }
  },
  "required": [
    "evidence_id"
  ],
  "additionalProperties": false
}