Omni Reader Multimodal Document Parser for Openclaw

Omni Reader connects an AI agent to a parsing service that turns documents, webpages, images, audio, and video into structured Markdown or clean text.

panting09266-ai
v1.2.14
Oct 12, 2026
0
889
0

Install & Download

1. ClawHub CLI

The fastest way to install a skill directly from the registry.

npx clawhub@latest install cue-omni-reader

2. Manual Installation

Copy the skill folder to one of these locations

Global
~/.openclaw/skills/
Workspace
<project>/skills/

Priority: Workspace > Local > Bundled

3. Prompt Installation

Copy this prompt to OpenClaw to install it automatically.

Help me install cue-omni-reader using Clawhub. If Clawhub is not installed, install it first (npm i -g clawhub).

Prefer to download?

Get the raw skill files in a ZIP archive.

What is Omni Reader Multimodal Document Parser?

Omni Reader is an Openclaw Skills integration guide for using the Cue Omni Reader MCP service to extract and organize content from local files and URLs. It supports PDFs, Office documents, images and scans, audio, video, webpages, text, code, and compressed archives, with outputs including Markdown, hypertext, or chunks where supported. OCR, speech recognition, visual analysis, and video keyframe understanding make it useful for more than conventional document conversion.

The skill describes how to connect through a local Node.js Bridge or a remote, URL-only MCP endpoint; parsing itself is performed by the Cue service, not locally by the skill. Privacy-conscious defaults include no_store=true, user-controlled API keys, and local-file access limited to authorized paths. For developers comparing Openclaw Skills for document extraction, transcription, or multimodal workflows, Omni Reader offers one consistent parsing interface and actionable guidance for handling asynchronous jobs, results, and errors.

Omni Reader Multimodal Document Parser Use Cases

  • Convert PDFs, Word documents, spreadsheets, presentations, scans, and images into Markdown for search, review, or downstream AI workflows.
  • Extract text from audio recordings using speech recognition, including multilingual meeting recordings.
  • Analyze video with speech recognition, keyframe visual understanding, and fused multimodal output; extract transcripts, subtitles, and on-screen text.
  • Parse a webpage directly from a URL while retaining useful document structure.
  • Process source code, structured text formats, CSV or Parquet data, and files inside supported archives.
  • Convert complex scanned tables into readable Markdown, then optionally export the result to Word, PDF, or Excel.
  • Connect an AI agent to an MCP service for repeatable parsing, status checks, cancellation, and—when using the local Bridge—result reading and export.

How Omni Reader Multimodal Document Parser Works

  1. Choose an input: provide an HTTP(S) URL or, with the local Bridge and user authorization, a local file path. Use the remote endpoint for URLs from anti-bot or login-dependent platforms; the remote endpoint cannot read local files.
  2. Select the appropriate MCP connection: install the local stdio Bridge for local-file access and local result-management tools, or configure the remote streamable-HTTP endpoint for URL-only use.
  3. Submit one parse request for the source. The default output is Markdown; supported alternatives include hypertext and chunks. The current supported detail setting is text.
  4. If parsing is asynchronous, retain the returned operation_id and poll get_parse_status after the indicated delay. Do not submit parse again to recover from a timeout or network interruption; that creates a new job and may incur another charge.
  5. Retrieve the completed result. Remote MCP responses include the result in the completed status response. The local Bridge can read large results in segments, read an outline, save a stable copy, or discard the artifact.
  6. Review and prepare the output for delivery. Convert HTML tables to readable Markdown when necessary, preserve merged-cell meaning, and ask before creating a separate Word, PDF, or Excel file.
  7. Handle structured errors according to their code and retryable flag. Retry status checks with the existing operation where possible, and only create a new parse operation when no usable operation remains.

Omni Reader Multimodal Document Parser Setup

  1. Create a Cue API key yourself at cuecue.cn. The agent cannot create or inspect your key. For the local Bridge, Node.js 20.12 or later is required.
  2. For local files or local result-management tools, install the Bridge:
npx -y @cueai/[email protected] setup
  1. Configure the local MCP server in your agent configuration, replacing the placeholders with your key and permitted local roots:
{
  "mcpServers": {
    "omni-reader": {
      "command": "npx",
      "args": ["-y", "@cueai/[email protected]"],
      "env": {
        "CUE_API_KEY": "<your-api-key>",
        "OMNI_ALLOWED_ROOTS": "<authorized-local-directories-separated-by-colons>"
      }
    }
  }
}
  1. For public or signed HTTPS URLs only, you can configure the remote MCP endpoint instead; it does not support local paths:
{
  "mcpServers": {
    "omni-reader": {
      "type": "streamable-http",
      "url": "https://mcp.cuecue.cn/api/omni-reader/mcp/",
      "headers": {
        "Authorization": "Bearer <your-api-key>"
      }
    }
  }
}
  1. Check the Bridge installation and configuration:
npx -y @cueai/[email protected] doctor --json
  1. Before parsing, run the lightweight version check when using the Bridge:
npx -y @cueai/[email protected] doctor --silent-check

Use a local path only after the agent obtains your confirmation. The Bridge is limited to user-authorized files and directories. Do not submit a bare oss:// URL; provide an accessible HTTP(S) or signed HTTPS URL instead. For Openclaw Skills setup, choose one connection mode based on the input and required tools.

Omni Reader Multimodal Document Parser Data Schema & Taxonomy

Inputs and limits

Input class Examples and handling
Documents PDF, Word, Excel, and PowerPoint; PDF support includes OCR, reading order, columns, and tables
Images and scans PNG, JPG, BMP, GIF, WebP, HEIC, and AVIF; OCR and visual understanding, including handwriting
Audio MP3, WAV, M4A, AAC, FLAC, and OGG; ASR transcription
Video MP4, MOV, MKV, WebM, AVI, and M4V; ASR, keyframe visual analysis, and multimodal fusion
Web and text HTTP(S) webpages, plain text, Markdown, code, JSON, YAML, TOML, XML, CSV, TSV, INI, LOG, and Parquet
Archives ZIP, RAR, TAR, GZ, TGZ, and BZ2

The documented limit is one file per parse request, with a maximum size of 256 MiB. Video is recommended in MP4 (H.264 + AAC); splitting long videos into 15–30 minute segments is advised. Default output is Markdown; hypertext and chunks are available through the documented remote output parameter. The currently supported detail value is text; grounded and layout are unavailable.

MCP tools and result lifecycle

Tool Availability Purpose
parse Remote and local Bridge Submit a URL or authorized local source for parsing
get_parse_status Remote and local Bridge Poll an operation and retrieve its status/result
cancel_parse Remote and local Bridge Cancel an in-progress parse
read_result Local Bridge only Read a result artifact in segments using result_id, cursor, and max_bytes
read_outline Local Bridge only Read the result outline
save_result Local Bridge only Export a stable result file
discard_result Local Bridge only Remove a result artifact

The remote endpoint exposes the first three tools; the local Bridge exposes all seven. Parsing can return an operation_id while processing. Preserve it and poll the same operation rather than resubmitting the source. MCP tool responses may wrap JSON in content[].text, which must be extracted and parsed before reading status fields.

Storage and metadata

  • Parsing defaults to no_store=true; the documentation states that source files and results are not stored on the server. Large results over 64 KiB may be temporarily cached locally and automatically cleared after 24 hours.
  • Local Bridge result artifacts can be read, saved to a stable file, or discarded. Results are not automatically described as permanent files; use save_result when a stable copy is required.
  • Structured errors can include code, failure_scope, retryable, user_action, and request_id. Use these fields to decide next steps and support troubleshooting.
  • The parse source accepts an HTTP(S) URL or an authorized local path. Use signed HTTPS links instead of bare oss:// URLs.

Omni Reader Multimodal Document Parser Advanced Features

  • Multimodal video parsing combines speech recognition, keyframe visual analysis, and transcript-aware content understanding; image and scan workflows also support OCR and visual interpretation.
  • Asynchronous job handling provides operation_id-based status polling, progress updates, and cancellation. Follow the returned polling delay, and never resubmit parse merely because a status request timed out.
  • The local Bridge adds segmented artifact reading, outline access, stable result export, and explicit cleanup; these tools are not available on the remote endpoint.
  • Input-based routing distinguishes local paths, ordinary public URLs, and platform URLs that require the remote endpoint. Local Bridge submissions of anti-bot platform URLs may be rejected.
  • Privacy controls include a default no-store setting, user-managed API keys, confirmation before local-file parsing, and configurable allowed roots for Bridge access.
  • Error handling is designed for recovery: respect retryable, retain the original operation where possible, apply backoff to status polling, and keep request_id for troubleshooting.
  • OCR and audio/video processing can report staged progress. Typical processing times depend on file type, length, layout complexity, and service load.
  • Markdown results can be converted with Pandoc to DOCX or PDF, or processed with Python pandas for Excel workflows. Convert raw HTML tables to readable Markdown before presenting them.
  • Openclaw Skills users can pair Omni Reader with platform-specific video skills when the goal extends beyond extraction to tasks such as script generation, video breakdowns, or study-note creation.

SKILL.md


Loading

Related Openclaw Skills

METADATA

Github Stars: 0
forks: 0

Featured*